bioinnovationbeacon.com

Category: AI in Biotechnology

  • AI unlocks genetic clues to personalize cancer treatment

    AI unlocks genetic clues to personalize cancer treatment

    A groundbreaking study led by USC Assistant Professor of Computer Science Ruishan Liu has uncovered how specific genetic mutations influence cancer treatment outcomes — insights that could help doctors tailor treatments more effectively. The largest study of its kind, the research analyzed data for more than 78,000 cancer patients across 20 cancer types. Patients received immunotherapies, chemotherapies and targeted therapies.

    Using advanced computational analysis, the researchers identified nearly 800 genetic changes that directly impacted survival outcomes. They also discovered 95 genes significantly associated with survival in cancers such as breast, ovarian, skin, and gastrointestinal cancers.

    Building on these insights, the team developed a machine learning tool to predict how patients with advanced lung cancer might respond to immunotherapy.

    “These discoveries highlight how genetic profiling can play a crucial role in personalizing cancer care,” said Liu. “By understanding how different mutations influence treatment response, doctors can select the most effective therapies — potentially avoiding ineffective therapies and focusing on those most likely to help.”

    Published in Nature Communications, the study highlights the critical roles of genes such as TP53, CDKN2A, and CDKN2B in influencing treatment outcomes, validating these associations with real-world data.

    Study co-authors are Shemra Rizzo, Lisa Wang, Nayan Chaudhary, Sophia Maund, and Sarah McGough and Ryan Copping of Genentech; Marius Rene Garmhausen of Roche; and James Zou of Stanford University.

    Why do mutations matter?

    Genetic mutations — changes in DNA — can influence how cancer develops and how a patient responds to treatment. Some mutations occur randomly, while others are inherited.

    In cancer, mutations can determine whether a tumor is more aggressive or how it might respond to certain treatments. Today, genetic testing is increasingly used in cancer care to identify these mutations, allowing doctors select treatments more precisely.

    For example, Patients diagnosed with non-small cell lung cancer (NSCLC) often receive genomic testing for mutations in genes like KRAS, EGFR and ALK to determine whether targeted therapies or immunotherapies might be effective.

    Key findings from the study include:

    • KRAS mutations in advanced non-small cell lung cancer were linked to poorer response to a common treatment (EGFR inhibitors), suggesting alternative treatments may be needed. • NF1 mutations improved responses to immunotherapy and worsened responses to certain targeted therapies, highlighting their complex role in treatment. • PI3K pathway mutations, which regulate cell growth, had varying effects depending on cancer type, with different responses in breast, melanoma and renal cancers. • DNA repair pathway mutations improved immunotherapy effectiveness in lung cancer by increasing tumor instability. • Mutations in immune-related pathways were associated with better survival rates for lung cancer patients treated with immunotherapy, suggesting not all mutations hinder treatment success.

    A powerful predictive tool

    While cancer treatments have traditionally followed a one-size-fits-all approach, where patients with the same type of cancer receive the same standard therapies, the study underscores the importance of precision medicine, which tailors treatment based on a patient’s unique genetic makeup.

    Yet while vast amounts of mutation data exist, only a small number have clinically validated treatments, limiting potential real-world impact and patient benefit. To bridge this gap, based on their findings, Liu’s team used machine learning to analyze how multiple mutations interact to influence treatment outcomes.

    “Our goal was to find patterns that might not be obvious at first glance.” Ruishan Liu

    They developed a Random Survival Forest (RSF) model, a predictive tool designed to refine treatment recommendations for lung cancer patients. By integrating large-scale real-world data with machine learning, the model identified new mutation-treatment interactions.

    “Our goal was to find patterns that might not be obvious at first glance, and then translate these insights into real-world tools that can expand access to immunotherapy for people with cancer,” Lui said. “One key innovation lies in integrating huge amounts of data with advanced statistical and machine learning techniques to uncover previously unrecognized mutation-treatment interactions.”

    While further clinical trials are needed, Liu sees this study as an important step toward making cancer treatment more precise and personalized.

    “This research shows the power of computational science in transforming complex clinical and genomic data into actionable insights,” she said. “It’s deeply fulfilling to contribute to tools and knowledge that can directly improve patient care.”

  • Transforming longevity research: AI paves the way for personalized treatments in aging science

    Transforming longevity research: AI paves the way for personalized treatments in aging science

    A collaborative study between researchers from the Yong Loo Lin School of Medicine, National University of Singapore (NUS Medicine), and the Institute for Biostatistics and Informatics in Medicine and Ageing Research, Rostock University Medical Center, Germany, investigated how advanced AI tools, like Large Language Models (LLMs), can make it easier to evaluate interventions for ageing and provide personalised recommendations. The findings were published in the leading review journal Ageing Research Reviews.

    Research into ageing is producing an overwhelming amount of data, making it difficult to determine which interventions — such as new medicines, dietary changes, or exercise routines — are safe and effective. This study investigated how AI can analyse data more efficiently and accurately, by proposing a comprehensive set of standards for AI systems to ensure they deliver accurate, reliable, and understandable evaluations through their ability to analyse complex biological data.

    The researchers identified eight critical requirements for effective AI-based evaluations:

    1. Correctness of the evaluation results. Data quality will be assessed for accuracy.
    2. Usefulness and comprehensiveness.
    3. Interpretability and explainability of the evaluation results. Clarity and conciseness of the results and the given explanations.
    4. Specific consideration of causal mechanisms affected by the intervention.
    5. Consideration of data in a holistic context:
      1. Efficacy and toxicity, and evidence for the existence of a large therapeutic window;
      2. Analyses in an “interdisciplinary” setting.
    6. Enabling reproducibility, standardisation, and harmonisation of the analyses (and of the reporting).
    7. Specific emphasis on diverse longitudinal large-scale data.
    8. Specific emphasis on results that relate to known mechanisms of ageing.

    Telling LLMs about these requirements as part of the prompting improved the quality of the recommendations they produced.

    Professor Brian Kennedy from the Department of Biochemistry & Physiology, and Healthy Longevity Translational Research Programme at NUS Medicine, who co-led the study, said, “We tested AI methods using real-world examples such as medicines and dietary supplements. We found that by following specific guidelines, AI can provide more accurate and detailed insights. For instance, when analysing rapamycin, a drug often studied for its potential to promote healthy ageing, the AI not only evaluated its efficacy but also provided context-specific explanations and caveats, such as possible side effects.”

    “The study’s findings could have far-reaching effects,” added Professor Georg Fuellen, Director, Institute for Biostatistics and Informatics in Medicine and Ageing Research, Rostock University Medical Center, who co-led the study, “For healthcare, telling the AI about the critical requirements of a good response can enable it to find more effective treatments and make them safer to use. Generally, AI tools could design better clinical trials and help tailor health recommendations to each person. This research is a major step toward using AI to improve health outcomes for everyone, especially as they age.”

    Moving forward, the team is now focusing on a large-scale study of how to best prompt AI models for longevity-related intervention advice, to evaluate their accuracy and reliability for a wide array of carefully designed benchmarks, that is, curated, high-quality data. The validation of such AI systems is specifically important because the longevity interventions may then be implemented by a large number of healthy people. Prospective studies will need to demonstrate that AI-based evaluations can accurately predict successful outcomes in human trials, paving the way for safer and more effective health interventions.

    The team hopes to use their findings to make health and longevity interventions more precise and accessible, and ultimately improve the quality and duration of life. Collaboration between researchers, clinicians, and policymakers will be essential to establish robust regulatory frameworks, ensuring the safe and effective use of AI-driven evaluations.

  • New tool enables phylogenomic analyses of entire genomes

    New tool enables phylogenomic analyses of entire genomes

    Researchers led by electrical engineers at the University of California San Diego have developed a better way to perform the comparative analysis of entire genomes. This approach can be used to study relationships between different species across geological time scales.

    This new approach is poised to unlock discoveries regarding how evolution has shaped present-day genomes and also how the tree of life is organized. The new method, named CASTER, is described in a paper published in Science on 23 January 2025.

    CASTER is poised to offer biologists a far more scalable approach than state-of-the-art for comparing full genomes. This is especially relevant given the exploding number of sequenced genomes of both living and also extinct species. All these genomes are available for comparative study through phylogenomic analyses. CASTER provides interpretable outputs that will help biologists understand not only the species relationships but also the mosaic of evolutionary histories across the genome.

    “Since the early 2000s, countless studies have claimed ‘genome-wide’ phylogeny reconstruction; however, these have been all based on subsampling regions scattered across the genomes, totalling only a small fraction of each full genome that is part of any given study. Analyzing all genomic positions while using complex models had seemed out of reach,” said Siavash Mirarab, an electrical engineering professor at UC San Diego and the corresponding author of the new paper in Science “What excites me is that we can now perform truly genome-wide analyses using every base pair aligned across species with widely available computational resources.”

    “Arriving at this milestone required combined efforts from many disciplines, including statistics, computer science, and biology,” said Chao Zhang, the first author on the new Science paper. Zhang completed his PhD at UC San Diego in the Bioinformatics and Systems Biology program. He is now on the faculty of the University of Copenhagen.

  • New method forecasts computation, energy costs for sustainable AI models

    New method forecasts computation, energy costs for sustainable AI models

    The process of updating deep learning/AI models when they face new tasks or must accommodate changes in data can have significant costs in terms of computational resources and energy consumption. Researchers have developed a novel method that predicts those costs, allowing users to make informed decisions about when to update AI models to improve AI sustainability.

    “There have been studies that focused on making deep learning model training more efficient,” says Jung-Eun Kim, corresponding author of a paper on the work and an assistant professor of computer science at North Carolina State University. “However, over a model’s life cycle, it will likely need to be updated many times. One reason is that, as our work here shows, retraining an existing model is much more cost effective than training a new model from scratch.

    “If we want to address sustainability issues related to deep learning AI, we must look at computational and energy costs across a model’s entire life cycle — including the costs associated with updates. If you cannot predict what the costs will be ahead of time, it is impossible to engage in the type of planning that makes sustainability efforts possible. That makes our work here particularly valuable.”

    Training a deep learning model is a computationally intensive process, and users want to go as long as possible without having to update the AI. However, two types of shifts can happen that make these updates inevitable. First, the task that the AI is performing may need to be modified. For example, if a model was initially tasked with only classifying digits and traffic symbols, you may need to modify the task to identify vehicles and humans as well. This is called a task shift.

    Second, the data users provide to the model may change. For example, you may need to make use of a new kind of data, or perhaps the data you are working with is being coded in a different way. Either way, the AI needs to be updated to accommodate the change. This is called a distribution shift.

    “Regardless of what is driving the need for an update, it is extremely useful for AI practitioners to have a realistic estimate of the computational demand that will be required for the update,” Kim says. “This can help them make informed decisions about when to conduct the update, as well as how much computational demand they will need to budget for the update.”

    To forecast what the computational and energy costs will be, the researchers developed a new technique they call the REpresentation Shift QUantifying Estimator (RESQUE).

    Essentially, RESQUE allows users to compare the dataset that a deep learning model was initially trained on to the new dataset that will be used to update the model. This comparison is done in a way that estimates the computational and energy costs associated with conducting the update.

    Those costs are presented as a single index value, which can then be compared with five metrics: epochs, parameter change, gradient norm, carbon and energy. Epochs, parameter change and gradient norm are all ways of measuring the amount of computational effort necessary to retrain the model.

    “However, to provide insight regarding what this means in a broader sustainability context, we also tell users how much energy, in kilowatt hours, will be needed to retrain the model,” Kim says. “And we predict how much carbon, in kilograms, will be released into the atmosphere in order to provide that energy.”

    The researchers conducted extensive experiments involving multiple data sets, many different distribution shifts, and many different task shifts to validate RESQUE’s performance.

    “We found that the RESQUE predictions aligned very closely with the real-world costs of conducting deep learning model updates,” Kim says. “Also, as I noted earlier, all of our experimental findings tell us that training a new model from scratch demands far more computational power and energy than retraining an existing model.”

    In the short term, RESQUE is a useful methodology for anyone who needs to update a deep learning model.

    “RESQUE can be used to help users budget computational resources for updates, allow them to predict how long the update will take, and so on,” Kim says.

    “In the bigger picture, this work offers a deeper understanding of the costs associated with deep learning models across their entire life cycle, which can help us make informed decisions related to the sustainability of the models and how they are used. Because if we want AI to be viable and useful, these models must be not only dynamic but sustainable.”

  • New AI predicts inner workings of cells

    New AI predicts inner workings of cells

    In the same way that ChatGPT understands human language, a new AI model developed by Columbia computational biologists captures the language of cells to accurately predict their activities.

    Using a new artificial intelligence method, researchers at Columbia University Vagelos College of Physicians and Surgeons can accurately predict the activity of genes within any human cell, essentially revealing the cell’s inner mechanisms. The system, described in the current issue of Nature, could transform the way scientists work to understand everything from cancer to genetic diseases.

    “Predictive generalizable computational models allow to uncover biological processes in a fast and accurate way. These methods can effectively conduct large-scale computational experiments, boosting and guiding traditional experimental approaches,” says Raul Rabadan, professor of systems biology and senior author of the new paper.

    Traditional research methods in biology are good at revealing how cells perform their jobs or react to disturbances. But they cannot make predictions about how cells work or how cells will react to change, like a cancer-causing mutation.

    “Having the ability to accurately predict a cell’s activities would transform our understanding of fundamental biological processes,” Rabadan says. “It would turn biology from a science that describes seemingly random processes into one that can predict the underlying systems that govern cell behavior.”

    In recent years, the accumulation of massive amounts of data from cells and more powerful AI models are starting to transform biology into a more predictive science. The 2024 Nobel Prize in Chemistry was awarded to researchers for their groundbreaking work in using AI to predict protein structures. But the use of AI methods to predict the activities of genes and proteins inside cells has proven more difficult.

    New AI method predicts gene expression in any cell

    In the new study, Rabadan and his colleagues tried to use AI to predict which genes are active within specific cells. Such information about gene expression can tell researchers the identity of the cell and how the cell performs its functions.

    “Previous models have been trained on data in particular cell types, usually cancer cell lines or something else that has little resemblance to normal cells,” Rabadan says. Xi Fu, a graduate student in Rabadan’s lab, decided to take a different approach, training a machine learning model on gene expression data from millions of cells obtained from normal human tissues. The inputs consisted of genome sequences and data showing which parts of the genome are accessible and expressed.

    The overall approach resembles the way ChatGPT and other popular “foundation” models work. These systems use a set of training data to identify underlying rules, the grammar of language, and then apply those inferred rules to new situations. “Here it’s exactly the same thing: we learn the grammar in many different cellular states, and then we go into a particular condition — it can be a diseased or it can be a normal cell type — and we can try to see how well we predict patterns from this information,” says Rabadan.

    Fu and Rabadan soon enlisted a team of collaborators, including co-first authors Alejandro Buendia, now a Stanford PhD student formerly in the Rabadan lab, and Shentong Mo of Carnegie Mellon, to train and test the new model.

    After training on data from more than 1.3 million human cells, the system became accurate enough to predict gene expression in cell types it had never seen, yielding results that agreed closely with experimental data.

    New AI methods reveal drivers of a pediatric cancer

    Next, the investigators showed the power of their AI system when they asked it to uncover still hidden biology of diseased cells, in this case, an inherited form of pediatric leukemia.

    “These kids inherit a gene that is mutated, and it was unclear exactly what it is these mutations are doing,” says Rabadan, who also co-directs the cancer genomics and epigenomics research program at Columbia’s Herbert Irving Comprehensive Cancer Center.

    With AI, the researchers predicted that the mutations disrupt the interaction between two different transcription factors that determine the fate of leukemic cells. Laboratory experiments confirmed AI’s prediction. Understanding the effect of these mutations uncovers specific mechanisms that drive this disease.

    AI could reveal “dark matter” in genome

    The new computational methods should also allow researchers to start exploring the role of genome’s “dark matter” — a term borrowed from cosmology that refers to the vast majority of the genome, which does not encode known genes — in cancer and other diseases.

    “The vast majority of mutations found in cancer patients are in so-called dark regions of the genome. These mutations do not affect the function of a protein and have remained mostly unexplored. says Rabadan. “The idea is that using these models, we can look at mutations and illuminate that part of the genome.”

    Already, Rabadan is working with researchers at Columbia and other universities, exploring different cancers, from brain to blood cancers, learning the grammar of regulation in normal cells, and how cells change in the process of cancer development.

    The work also opens new avenues for understanding many diseases beyond cancer and potentially identifying targets for new treatments. By presenting novel mutations to the computer model, researchers can now gain deep insights and predictions about exactly how those mutations affect a cell.

    Coming on the heels of other recent advances in artificial intelligence for biology, Rabadan sees the work as part of a major trend: “It’s really a new era in biology that is extremely exciting; transforming biology into a predictive science.”

  • How good are AI doctors at medical conversations?

    How good are AI doctors at medical conversations?

    Artificial intelligence tools such as ChatGPT have been touted for their promise to alleviate clinician workload by triaging patients, taking medical histories and even providing preliminary diagnoses.

    These tools, known as large-language models, are already being used by patients to make sense of their symptoms and medical tests results.

    But while these AI models perform impressively on standardized medical tests, how well do they fare in situations that more closely mimic the real world?

    Not that great, according to the findings of a new study led by researchers at Harvard Medical School and Stanford University.

    For their analysis, published Jan. 2 in Nature Medicine, the researchers designed an evaluation framework — or a test — called CRAFT-MD (Conversational Reasoning Assessment Framework for Testing in Medicine) and deployed it on four large-language models to see how well they performed in settings closely mimicking actual interactions with patients.

    All four large-language models did well on medical exam-style questions, but their performance worsened when engaged in conversations more closely mimicking real-world interactions.

    This gap, the researchers said, underscores a two-fold need: First, to create more realistic evaluations that better gauge the fitness of clinical AI models for use in the real world and, second, to improve the ability of these tools to make diagnosis based on more realistic interactions before they are deployed in the clinic.

    Evaluation tools like CRAFT-MD, the research team said, can not only assess AI models more accurately for real-world fitness but could also help optimize their performance in clinic.

    “Our work reveals a striking paradox — while these AI models excel at medical board exams, they struggle with the basic back-and-forth of a doctor’s visit,” said study senior author Pranav Rajpurkar, assistant professor of biomedical informatics at Harvard Medical School. “The dynamic nature of medical conversations — the need to ask the right questions at the right time, to piece together scattered information, and to reason through symptoms — poses unique challenges that go far beyond answering multiple choice questions. When we switch from standardized tests to these natural conversations, even the most sophisticated AI models show significant drops in diagnostic accuracy.”

    A better test to check AI’s real-world performance

    Right now, developers test the performance of AI models by asking them to answer multiple choice medical questions, typically derived from the national exam for graduating medical students or from tests given to medical residents as part of their certification.

    “This approach assumes that all relevant information is presented clearly and concisely, often with medical terminology or buzzwords that simplify the diagnostic process, but in the real world this process is far messier,” said study co-first author Shreya Johri, a doctoral student in the Rajpurkar Lab at Harvard Medical School. “We need a testing framework that reflects reality better and is, therefore, better at predicting how well a model would perform.”

    CRAFT-MD was designed to be one such more realistic gauge.

    To simulate real-world interactions, CRAFT-MD evaluates how well large-language models can collect information about symptoms, medications, and family history and then make a diagnosis. An AI agent is used to pose as a patient, answering questions in a conversational, natural style. Another AI agent grades the accuracy of final diagnosis rendered by the large-language model. Human experts then evaluate the outcomes of each encounter for ability to gather relevant patient information, diagnostic accuracy when presented with scattered information, and for adherence to prompts.

    The researchers used CRAFT-MD to test four AI models — both proprietary or commercial and open-source ones — for performance in 2,000 clinical vignettes featuring conditions common in primary care and across 12 medical specialties.

    All AI models showed limitations, particularly in their ability to conduct clinical conversations and reason based on information given by patients. That, in turn, compromised their ability to take medical histories and render appropriate diagnosis. For example, the models often struggled to ask the right questions to gather pertinent patient history, missed critical information during history taking, and had difficulty synthesizing scattered information. The accuracy of these models declined when they were presented with open-ended information rather than multiple choice answers. These models also performed worse when engaged in back-and-forth exchanges — as most real-world conversations are — rather than when engaged in summarized conversations.

    Recommendations for optimizing AI’s real-world performance

    Based on these findings, the team offers a set of recommendations both for AI developers who design AI models and for regulators charged with evaluating and approving these tools.

    These include:

    • Use of conversational, open-ended questions that more accurately mirror unstructured doctor-patient interactions in the design, training, and testing of AI tools
    • Assessing models for their ability to ask the right questions and to extract the most essential information
    • Designing models capable of following multiple conversations and integrating information from them
    • Designing AI models capable of integrating textual (notes from conversations) with and non-textual data (images, EKGs)
    • Designing more sophisticated AI agents that can interpret non-verbal cues such as facial expressions, tone, and body language

    Additionally, the evaluation should include both AI agents and human experts, the researchers recommend, because relying solely on human experts is labor-intensive and expensive. For example, CRAFT-MD outpaced human evaluators, processing 10,000 conversations in 48 to 72 hours, plus 15-16 hours of expert evaluation. In contrast, human-based approaches would require extensive recruitment and an estimated 500 hours for patient simulations (nearly 3 minutes per conversation) and about 650 hours for expert evaluations (nearly 4 minutes per conversation). Using AI evaluators as first line has the added advantage of eliminating the risk of exposing real patients to unverified AI tools.

    The researchers said they expect that CRAFT-MD itself will also be updated and optimized periodically to integrate improved patient-AI models.

    “As a physician scientist, I am interested in AI models that can augment clinical practice effectively and ethically,” said study co-senior author Roxana Daneshjou, assistant professor of Biomedical Data Science and Dermatology at Stanford University. “CRAFT-MD creates a framework that more closely mirrors real-world interactions and thus it helps move the field forward when it comes to testing AI model performance in health care.”

    Authorship, funding, disclosures

    Additional authors included Jaehwan Jeong and Hong-Yu Zhou, Harvard Medical School; Benjamin A. Tran, Georgetown University; Daniel I. Schlessinger, Northwestern University; Shannon Wongvibulsin, University of California-Los Angeles; Leandra A. Barnes, Zhuo Ran Cai and David Kim, Sandford University; and Eliezer M. Van Allen, Dana-Farber Cancer Institute.

    The work was supported by the HMS Dean’s Innovation Award and a Microsoft Accelerate Foundation Models Research grant awarded to Pranav Rajpurkar. SJ received further support through the IIE Quad Fellowship.

    Daneshjou reported receiving personal fees from DWA, personal fees from Pfizer, personal fees from L’Oreal, personal fees from VisualDx, stock options from MDAlgorithms and Revea outside the submitted work, and a patent for TrueImage pending. Schlessinger is the co-founder of FixMySkin Healing Balms, a shareholder in Appiell Inc. and K-Health, a consultant with Appiell Inc and LuminDx, and an investigator for Abbvie and Sanofi. Van Allen serves as an advisor to Enara Bio, Manifold Bio, Monte Rosa, Novartis Institute for Biomedical Research, Serinus Bio. E.M.V.A provides research support to Novartis, BMS, Sanofi, NextPoint. Van Allen holds equity in Tango Therapeutics, Genome Medical, Genomic Life, Enara Bio, Manifold Bio, Microsoft, Monte Rosa, Riva Therapeutics, Serinus Bio, Syapse. Van Allen has filed for institutional patents on chromatin mutations and immunotherapy response, and methods for clinical interpretation; intermittent legal consulting on patents for Foaley & Hoag, and serves on the editorial board of Science Advances.

  • Fast, rewritable computing with DNA origami registers

    Fast, rewritable computing with DNA origami registers

    DNA stores the instructions for life and, along with enzymes and other molecules, computes everything from hair color to risk of developing diseases. Harnessing that prowess and immense storage capacity could lead to DNA-based computers that are faster and smaller than today’s silicon-based versions. As a step toward that goal, researchers report in ACS Central Science a fast, sequential DNA computing method that is also rewritable — just like current computers.

    “DNA computing as a liquid computing paradigm has unique application scenarios and offers the potential for massive data storage and processing of digital files stored in DNA,” says Fei Wang, a co-author of the study.

    In living organisms, DNA expression occurs sequentially: Genes are transcribed into RNA, which is translated into proteins. This process happens to many genes simultaneously and repeatedly. If researchers can duplicate this complex, elegant dance in DNA-based computers, these devices could be more powerful than current silicon-based machines. Researchers have demonstrated sequential DNA computing for very focused, specialized tasks. But until recently, not much progress had been made in developing more general and programmable DNA devices that could be used and reused for various applications.

    In previous research, Chunhai Fan, Wang and colleagues developed a programmable DNA integrated circuit with many logic gates that act as instructions for the circuit’s operations. Here’s how it worked:

    • Data, 0 or 1, was represented by a short piece of single-stranded DNA, called an oligonucleotide, that contained a series of bases: adenine, thymine, guanine and cytosine. (In nature, the sequence of bases codes for a gene.)
    • For example, two inputs of 1 (DNA strands 1 and 2) would interact with an OR logic gate DNA molecule.
    • Then in a fluid-filled tube, the input oligonucleotide interacted with a logic gate DNA molecule and generated an output oligonucleotide.
    • The output oligonucleotide bound to a different single-stranded DNA that was folded into an origami-like structure, called a register in computer lingo.
    • The oligonucleotide was “read” by reviewing its base sequence, released and used in a vial containing the next gate, and so on.

    This process took hours, and someone had to manually transfer the oligonucleotide from one gate to another vial for the next computing operation. So the team, along with Hui Lv and Sisi Jia, wanted to speed things up.

    To make the reaction processes more efficient and compact, the team first placed the DNA origami register onto a solid glass 2D surface. The output oligonucleotide floating in liquid from a specific logic gate then attached to the glass-mounted register. After the output oligonucleotide was read and the logic gate instructions determined, it detached, which reset the register so it could be rewritten, thereby avoiding the need to move or replace registers. The researchers also designed an amplifier that boosted the output signal so all the pieces — the gates, oligonucleotides and registers — could find one another more easily. In a proof-of-concept experiment, all the DNA computing reactions took place in a single tube within 90 minutes.

    “This research paves the way for developing large-scale DNA computing circuits with high speed and lays the foundation for visual debugging and automated execution of DNA molecular algorithms,” says Wang.

  • Better environmental performance boosts profits and cuts costs

    Better environmental performance boosts profits and cuts costs

    Sustainable practices in business are more than just an ethical responsibility; they make sound financial sense. Researchers from Kyushu University, in a study published on December 10, 2024, in Corporate Social Responsibility and Environmental Management, reveal that companies with better environmental performance and transparent disclosures can lower costs and boost profits.

    Investors are increasingly recognizing companies’ contributions toward carbon neutrality, driving the growth of environmental, social, and governance (ESG) investing. To support this trend, the Sustainability Accounting Standards Board (SASB) has provided an industry-specific framework to help companies effectively communicate their sustainability risks and opportunities to investors. Many companies now disclose their environmental information using this framework, and in many countries around the world, such transparency is becoming mandatory.

    Despite the advancements, the impact of corporate environmental strategies and performance on costs and profits remains unclear. To address this, Professor Hidemichi Fujii from Kyushu University’s Faculty of Economics and his team analyzed financial and environmental data from 8,547 companies across 34 countries spanning 2015 to 2022.

    The team developed two quantitative indicators to assess corporate environmental information: materiality-based scores and overall environmental scores.

    “Financial materiality is a relatively new concept. Environmental priorities vary across industries, as different companies face different key environmental challenges. Financial materiality helps investors assess whether disclosed information is relevant, and supports informed decision-making,” explained Siyu Shen, a graduate student at Kyushu University’s Graduate School of Economics and the paper’s first author.

    According to the SASB framework, environmental issues can be categorized into six areas, including greenhouse gas emissions, and water & wastewater management. For instance, water management is highly relevant to industries like mining but less critical for sectors such as finance. Materiality-based scores quantify only the relevant issues to see how efficiently a company addresses environmental challenges, while overall environmental scores evaluate all disclosed information to assess a company’s general environmental efforts.

    The researchers applied these two scores to assess companies’ environmental disclosures and performance. They found that companies with stronger environmental engagement can achieve better financial outcomes, including enhanced short- and long-term profits, and reduced costs. Notably, firms with superior environmental performance — rather than those focusing merely on disclosure — demonstrate better financial results and attract greater interest from investors.

    “Investors value what companies do for the environment more than what they say,” says Fujii. “By taking concrete action on environmental issues, companies signal sustainability and reliability to consumers and investors, lowering perceived risks, and strengthening their appeal as stable and ethical investments.”

    While overall environmental scores have a clear positive link to financial performance, materiality-based scores show only a limited correlation. This contradicted the team’s hypothesis, leading them to explore differences in how environmental efficiency is valued across countries.

    A closer look at the global data reveals that environmental efficiency is more strongly tied to financial performance in developed countries, such as America and Japan. In contrast, it remains less significant in developing countries like Chile and Indonesia.

    “This difference likely reflects variations in environmental regulations and public awareness across countries,” Shen explains. “In more economically developed countries, where companies have long been engaged in sustainability efforts, improving environmental efficiency can enhance profitability and market valuation. Meanwhile, in developing regions, as the overall regulatory frameworks are still developing, the priority is placed on environmental performance and transparency rather than efficiency.”

    The team is further investigating how macroeconomic factors, such as regulatory and social environments,influence corporate sustainability practices and financial outcomes across countries. Through a series of studies, they aim to provide scientific evidence of the impact of corporate environmental information disclosure and conservation efforts on economic performance. “We expect our international comparative studies to offer useful information for promoting effective policy planning to promote proactive responses to environmental issues,” Fujii adds.

  • Analyzing multiple mammograms improves breast cancer risk prediction

    Analyzing multiple mammograms improves breast cancer risk prediction

    A new study from Washington University School of Medicine in St. Louis describes an innovative method of analyzing mammograms that significantly improves the accuracy of predicting the risk of breast cancer development over the following five years. Using up to three years of previous mammograms, the new method identified individuals at high risk of developing breast cancer 2.3 times more accurately than the standard method, which is based on questionnaires assessing clinical risk factors alone, such as age, race and family history of breast cancer.

    The study is published Dec. 5 in JCO Clinical Cancer Informatics.

    “We are seeking ways to improve early detection, since that increases the chances of successful treatment,” said senior author Graham A. Colditz, MD, DrPH, associate director of Siteman Cancer Center, based at Barnes-Jewish Hospital and WashU Medicine, and the Niess-Gain Professor of Surgery. “This improved prediction of risk also may help research surrounding prevention, so that we can find better ways for women who fall into the high-risk category to lower their five-year risk of developing breast cancer.”

    This risk-prediction method builds on past research led by Colditz and lead author Shu (Joy) Jiang, PhD, a statistician, data scientist and associate professor of surgery in the Division of Public Health Sciences at WashU Medicine. The researchers showed that prior mammograms hold a wealth of information on early signs of breast cancer development that can’t be perceived even by a well-trained human eye. This information includes subtle changes over time in breast density, which is a measure of the relative amounts of fibrous versus fatty tissue in the breasts.

    For the new study, the team built an algorithm based on artificial intelligence that can discern subtle differences in mammograms and help identify those women at highest risk of developing a new breast tumor over a specific timeframe. In addition to breast density, their machine-learning tool considers changes in other patterns in the images, including in texture, calcification and asymmetry within the breasts.

    “Our new method is able to detect subtle changes over time in repeated mammogram images that are not visible to the eye,” said Jiang, yet these changes hold rich information that can help identify high-risk individuals.

    At the moment, risk-reduction options are limited and can include drugs such as tamoxifen that lower risk but may have unwanted side effects. Most of the time, women at high risk are offered more frequent screening or the option of adding another imaging method, such as an MRI, to try to identify cancer as early as possible.

    “Today, we don’t have a way to know who is likely to develop breast cancer in the future based on their mammogram images,” said co-author Debbie L. Bennett, MD, an associate professor of radiology and chief of breast imaging for the Mallinckrodt Institute of Radiology at WashU Medicine. “What’s so exciting about this research is that it indicates that it is possible to glean this information from current and prior mammograms using this algorithm. The prediction is never going to be perfect, but this study suggests the new algorithm is much better than our current methods.”

    AI improves prediction of breast cancer development

    The researchers trained their machine-learning algorithm on the mammograms of more than 10,000 women who received breast cancer screenings through Siteman Cancer Center from 2008 — 2012. These individuals were followed through 2020, and in that time 478 were diagnosed with breast cancer.

    The researchers then applied their method to predict breast cancer risk in a separate set of patients — more than 18,000 women who received mammograms through Emory University in the Atlanta area from 2013 — 2020. Subsequently, 332 women were diagnosed with breast cancer during the follow-up period, which ended in 2020.

    According to the new prediction model, women in the high-risk group were 21 times more likely to be diagnosed with breast cancer over the following five years than were those in the lowest-risk group. In the high-risk group, 53 out of every 1,000 women screened developed breast cancer over the next five years. In contrast, in the low-risk group, 2.6 women per 1,000 screened developed breast cancer over the following five years. Under the old questionnaire-based methods, only 23 women per 1,000 screened were correctly classified in the high-risk group, providing evidence that the old method, in this case, missed 30 breast cancer cases that the new method found.

    The mammograms were conducted at academic medical centers and community clinics, demonstrating that the accuracy of the method holds up in diverse settings. Importantly, the algorithm was built with robust representation of Black women, who are usually underrepresented in development of breast cancer risk models. The accuracy for predicting risk held up across racial groups. Of the women screened through Siteman, most were white, and 27% were Black. Of those screened through Emory, 42% were Black.

    In ongoing work, the researchers are testing the algorithm in women of diverse racial and ethnic backgrounds, including those of Asian, southeast Asian and Native American descent, to help ensure that the method is equally accurate for everyone.

    The researchers are working with WashU’s Office of Technology Management toward patents and licensing on the new method with the goal of making it broadly available anywhere screening mammograms are provided. Colditz and Jiang also are working toward founding a start-up company around this technology.

    Jiang S, Bennett DL, Rosner BA, Tamimi RM, Colditz GA. Development and validation of a dynamic 5-year breast cancer risk model using repeated mammograms. JCO Clinical Cancer Informatics. Dec. 5, 2024.

    This work was supported by Washington University School of Medicine in St. Louis.

    Jiang and Colditz have patents pending related to this work, predicting disease risk using radiomic images.

  • Persistent problems with AI-assisted genomic studies

    Persistent problems with AI-assisted genomic studies

    University of Wisconsin-Madison researchers are warning that artificial intelligence tools gaining popularity in the fields of genetics and medicine can lead to flawed conclusions about the connection between genes and physical characteristics, including risk factors for diseases like diabetes.

    The faulty predictions are linked to researchers’ use of AI to assist genome-wide association studies. Such studies scan through hundreds of thousands of genetic variations across many people to hunt for links between genes and physical traits. Of particular interest are possible connections between genetic variations and certain diseases.

    Genetics’ link to disease not always straightforward

    Genetics play a role in the development of many health conditions. While changes in some individual genes are directly connected to an increased risk for diseases like cystic fibrosis, the relationship between genetics and physical traits is often more complicated.

    Genome-wide association studies have helped to untangle some of these complexities, often using large databases of individuals’ genetic profiles and health characteristics, such as the National Institutes of Health’s All of Us project and the UK Biobank. However, these databases are often missing data about health conditions that researchers are trying to study.

    “Some characteristics are either very expensive or labor-intensive to measure, so you simply don’t have enough samples to make meaningful statistical conclusions about their association with genetics,” says Qiongshi Lu, an associate professor in the UW-Madison Department of Biostatistics and Medical Informatics and an expert on genome-wide association studies.

    The risks of bridging data gaps with AI

    Researchers are increasingly attempting to work around this problem by bridging data gaps with ever more sophisticated AI tools.

    “It has become very popular in recent years to leverage advances in machine learning, so we now have these advanced machine-learning AI models that researchers use to predict complex traits and disease risks with even limited data,” Lu says.

    Now, Lu and his colleagues have demonstrated the peril of relying on these models without also guarding against biases they may introduce. The team describe the problem in a paper recently published in the journal Nature Genetics. In it, Lu and his colleagues show that a common type of machine learning algorithm employed in genome-wide association studies can mistakenly link several genetic variations with an individual’s risk for developing Type 2 diabetes.

    “The problem is if you trust the machine learning-predicted diabetes risk as the actual risk, you would think all those genetic variations are correlated with actual diabetes even though they aren’t,” says Lu.

    These “false positives” are not limited to these specific variations and diabetes risk, Lu adds, but are a pervasive bias in AI-assisted studies.

    New statistical method can reduce false positives

    In addition to identifying the problem with overreliance on AI tools, Lu and his colleagues propose a statistical method that researchers can use to guarantee the reliability of their AI-assisted genome-wide association studies. The method helps removing bias that machine learning algorithms can introduce when they’re making inferences based on incomplete information.

    “This new strategy is statistically optimal,” Lu says, noting that the team used it to better pinpoint genetic associations with individuals’ bone mineral density.

    AI not the only problem with some genome-wide association studies

    While the group’s proposed statistical method could help improve the accuracy of AI-assisted studies, Lu and his colleagues also recently identified problems with similar studies that fill data gaps with proxy information rather than algorithms.

    In another recently published paper appearing in Nature Genetics, the researchers ring the alarm about studies that over-rely on proxy information in an attempt to establish connections between genetics and certain diseases.

    For instance, large health databases like the UK Biobank have a ton of genetic information about large populations, but they don’t have very much data regarding the incidence of diseases that tend to crop up later in life, like most neurodegenerative diseases.

    For Alzheimer’s disease specifically, some researchers have attempted to bridge that gap with proxy data gathered through family health history surveys, where individuals can report a parent’s Alzheimer’s diagnosis.

    The UW-Madison team found that such proxy-information studies can produce “highly misleading genetic correlation” between Alzheimer’s risk and higher cognitive abilities.

    “These days, genomic scientists routinely work with biobank datasets that have hundreds of thousands of individuals, however, as statistical power goes up, biases and the probability of errors are also amplified in these massive datasets,” says Lu. “Our group’s recent studies provide humbling examples and highlight the importance of statistical rigor in biobank-scale research studies.”