About

Russ Altman, MD, PhD, is a professor associated with the Stanford Helix Group, specializing in Biomedical Data Science and Medicine. His research focuses on biomedical informatics, integrating data science with medical research to advance understanding and treatment of health conditions. As a principal investigator, he contributes to the development of innovative approaches in biomedical data analysis, leveraging his extensive background in medicine and biomedical informatics to address complex biological and clinical questions.

Research topics

• Computer Science
• Artificial Intelligence
• Political Science
• Genetics
• Law
• Engineering
• Machine Learning
• Biology
• Data science
• Medicine
• Medical physics
• Psychology
• Computational biology
• Algorithm
• Pharmacology
• Management science
• Bioinformatics
• Systems engineering
• Engineering ethics

Selected publications

• Large language models can disambiguate opioid slang on social media

  ArXiv.org · 2026-03-11

  articleOpen accessSenior author

  Social media text shows promise for monitoring trends in the opioid overdose crisis; however, the overwhelming majority of social media text is unrelated to opioids. When leveraging social media text to monitor trends in the ongoing opioid overdose crisis, a common strategy for identifying relevant content is to use a lexicon of opioid-related terms as inclusion criteria. However, many slang terms for opioids, such as "smack" or "blues," have common non-opioid meanings, making them ambiguous. The advanced textual reasoning capability of large language models (LLMs) presents an opportunity to disambiguate these slang terms at scale. We present three tasks on which to evaluate four state-of-the-art LLMs (GPT-4, GPT-5, Gemini 2.5 Pro, and Claude Sonnet 4.5): a lexicon-based setting, in which the LLM must disambiguate a specific term within the context of a given post; a lexicon-free setting, in which the LLM must identify opioid-related posts from context without a lexicon; and an emergent slang setting, in which the LLM must identify opioid-related posts with simulated new slang terms. All four LLMs showed excellent performance across all tasks. In both subtasks of the lexicon-based setting, LLM F1 scores ("fenty" subtask: 0.824-0.972; "smack" subtask: 0.540-0.862) far exceeded those of the best lexicon strategy (0.126 and 0.009, respectively). In the lexicon-free task, LLM F1 scores (0.544-0.769) surpassed those of lexicons (0.080-0.540), and LLMs demonstrated uniformly higher recall. On emergent slang, all LLMs had higher accuracy (average: 0.784), F1 score (average: 0.712), precision (average: 0.981), and recall (average: 0.587) than the two lexicons assessed. Our results show that LLMs can be used to identify relevant content for low-prevalence topics, including but not limited to opioid references, enhancing data provided to downstream analyses and predictive models.

  PublisherOA PDF

• TikTok is a valuable data source for tracking the opioid crisis

  npj Digital Medicine · 2026-04-27

  articleOpen accessSenior author

  Monitoring opioid-related chatter on social media can predict the course of opioid addiction and the overdose epidemic. We assessed the utility of TikTok, a prominent short video-based social media platform, as a means of tracking the opioid addiction and overdose crisis. We collected 569,581 TikTok comments (posted between January 2021 and June 2025) from 48,306 opioid-related videos, making this study the first large-scale analysis of TikTok comments for opioid surveillance. We extracted 200 topics from these comments using Latent Dirichlet Allocation (LDA) and incorporated the topics into ARIMA models that forecast synthetic opioid mortality over 6-month horizons. We also analyzed conversational patterns using the LIWC2015 pronoun dictionaries and GPT o1-mini. We found that (1) incorporating TikTok topics into the ARIMA models reduced forecasting Mean Absolute Error by up to 37% (2) the topics spanned five broad themes (use, source, recovery, harm-reduction, loss), showing the diversity of opioid discourse on TikTok, and (3) TikTok comments included first-person, second-person, and third-person accounts of opioid use (i.e., personal use, engaging with other users in conversation about their use, and relating views of others' use, respectively). These findings emphasize the usefulness of TikTok comments as a data source for opioid use surveillance.

  PublisherOA PDFDOI

• Large language models can disambiguate opioid slang on social media

  arXiv (Cornell University) · 2026-03-11

  preprintOpen accessSenior author

  Social media text shows promise for monitoring trends in the opioid overdose crisis; however, the overwhelming majority of social media text is unrelated to opioids. When leveraging social media text to monitor trends in the ongoing opioid overdose crisis, a common strategy for identifying relevant content is to use a lexicon of opioid-related terms as inclusion criteria. However, many slang terms for opioids, such as "smack" or "blues," have common non-opioid meanings, making them ambiguous. The advanced textual reasoning capability of large language models (LLMs) presents an opportunity to disambiguate these slang terms at scale. We present three tasks on which to evaluate four state-of-the-art LLMs (GPT-4, GPT-5, Gemini 2.5 Pro, and Claude Sonnet 4.5): a lexicon-based setting, in which the LLM must disambiguate a specific term within the context of a given post; a lexicon-free setting, in which the LLM must identify opioid-related posts from context without a lexicon; and an emergent slang setting, in which the LLM must identify opioid-related posts with simulated new slang terms. All four LLMs showed excellent performance across all tasks. In both subtasks of the lexicon-based setting, LLM F1 scores ("fenty" subtask: 0.824-0.972; "smack" subtask: 0.540-0.862) far exceeded those of the best lexicon strategy (0.126 and 0.009, respectively). In the lexicon-free task, LLM F1 scores (0.544-0.769) surpassed those of lexicons (0.080-0.540), and LLMs demonstrated uniformly higher recall. On emergent slang, all LLMs had higher accuracy (average: 0.784), F1 score (average: 0.712), precision (average: 0.981), and recall (average: 0.587) than the two lexicons assessed. Our results show that LLMs can be used to identify relevant content for low-prevalence topics, including but not limited to opioid references, enhancing data provided to downstream analyses and predictive models.

  PublisherDOI

• Paying attention to attention: High attention sites as indicators of protein family and function in language models

  PLoS Computational Biology · 2025-09-12 · 4 citations

  articleOpen accessSenior authorCorresponding

  Protein Language Models (PLMs) use transformer architectures to capture patterns within protein primary sequences, providing a powerful computational representation of the amino acid sequence. Through large-scale training on protein primary sequences, PLMs generate vector representations that encapsulate the biochemical and structural properties of proteins. At the core of PLMs is the attention mechanism, which facilitates the capture of long-range dependencies by computing pairwise importance scores across residues, thereby highlighting regions of biological interaction within the sequence. The attention matrices offer an untapped opportunity to uncover specific biological properties of proteins, particularly their functions. In this work, we introduce a novel approach, using the Evolutionary Scale Modelling (ESM), for identifying High Attention (HA) sites within protein primary sequences, corresponding to key residues that define protein families. By examining attention patterns across multiple layers, we pinpoint residues that contribute most to family classification and function prediction. Our contributions are as follows: (1) we propose a method for identifying HA sites at critical residues from the middle layers of the PLM; (2) we demonstrate that these HA sites provide interpretable links to biological functions; and (3) we show that HA sites improve active site predictions for functions of unannotated proteins. We make available the HA sites for the human proteome. This work offers a broadly applicable approach to protein classification and functional annotation and provides a biological interpretation of the PLM's representation.

  PublisherDOI

• The E2F4 transcriptional repressor is a key mechanistic regulator of colon cancer resistance to <i>irinotecan</i> (CPT-11)

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-01-24 · 2 citations

  preprintOpen access

  ABSTRACT Background. Colorectal carcinomas (CRCs) are seldom eradicated by cytotoxic chemotherapy. Cancer cells with stem-like functional properties, often referred to as “cancer stem cells” (CSCs), display preferential resistance to several anti-tumor agents used in cancer chemotherapy, but the molecular mechanisms underpinning their selective survival remain only partially understood. Methods. In this study, we used Transcription Factor Target Genes (TFTG) enrichment analysis to identify transcriptional regulators (activators or repressors) that undergo preferential activation by chemotherapy in CRC cells with a “bottom-of-the-crypt” phenotype (EPCAM + /CD44 + /CD166 + ; CSC-enriched) as compared to CRC cells with a “top-of-the-crypt” phenotype (EPCAM + /CD44 neg /CD166 neg ; CSC-depleted). The two cell populations were purified in parallel by fluorescence-activated cell sorting (FACS) from a patient-derived xenograft (PDX) line representative of a moderately differentiated human CRC, following in vivo chemotherapy with irinotecan (CPT-11). The transcriptional regulators identified as differentially activated were tested for differential expression in normal vs. cancer tissues, and in cell populations enriched in stem/progenitor cell-types as compared to differentiated lineages (goblet cells, enterocytes) in the mouse colon epithelium. Finally, the top candidate was tested for mechanistic contribution to drug-resistance by selective down-regulation using short-hairpin RNAs (shRNAs). Results. Our analysis identified E2F4 and TFDP1, two core components of the DREAM transcriptional repression complex, as transcriptional modulators preferentially activated by irinotecan in EPCAM + /CD44 + /CD166 + as compared to EPCAM + /CD44 neg /CD166 neg cancer cells. The expression levels of both genes ( E2F4 , TFDP1 ) were found up-regulated in CRCs as compared to human normal colon tissues, and in a sub-population of mouse colon epithelial cells enriched in stem/progenitor elements (Epcam + /Cd44 + /Cd66a low /Kit neg ) as compared to other sub-populations enriched in either goblet cells (Epcam + /Cd44 + /Cd66a low /Kit + ) or enterocytes (Epcam + /Cd44 neg /Cd66a high ). Most importantly, E2F4 down-regulation using shRNAs dramatically enhanced the sensitivity of human CRCs to in vivo treatment with irinotecan , across three independent PDX models. Conclusions. Our data identified E2F4 and the DREAM repressor complex as critical regulators of human CRC resistance to irinotecan , and as candidate targets for the development of chemo-sensitizing agents.

  PublisherDOI

• Empirical Drug Dosage Validates Pharmacogenomic Associations in All of Us

  Clinical and Translational Science · 2025-11-01

  articleOpen accessSenior authorCorresponding

  The All of Us research program, a national longitudinal study conducted by the US National Institutes of Health, provides robust medical history, drug dosage and genomic data from a diverse population. All of Us offers an opportunity to discover novel correlations between drug dosage and genetic variation. However, first it is necessary to evaluate the quality and quantity of the data and its ability to replicate known associations. In this paper, we investigate whether known drug-gene interactions can be recovered from the All of Us dataset, based on data from electronic health records. Focusing on the Cytochrome P450 (CYP450) enzyme family, which metabolizes approximately 90% of clinically available drugs, we evaluate 61 drugs metabolized by the enzymes. We then identify significant differences in drug dosages across CYP450 metabolizer phenotypes. Our results validate some known interactors of CYP2D6, CYP2C19, CYP2C9, and CYP3A5. However, we did not recover all validated PGx interactions, potentially due to noise, lack of doctors adjusting drug dosage or phenoconversion. Nevertheless, our findings highlight the potential of the All of Us dataset, which captures some known pharmacogenomic interactions.

  PublisherOA PDFDOI

• CRISPR-GPT for agentic automation of gene-editing experiments

  Nature Biomedical Engineering · 2025-07-30 · 53 citations

  articleOpen access

  Performing effective gene-editing experiments requires a deep understanding of both the CRISPR technology and the biological system involved. Meanwhile, despite their versatility and promise, large language models (LLMs) often lack domain-specific knowledge and struggle to accurately solve biological design problems. We present CRISPR-GPT, an LLM agent system to automate and enhance CRISPR-based gene-editing design and data analysis. CRISPR-GPT leverages the reasoning capabilities of LLMs for complex task decomposition, decision-making and interactive human-artificial intelligence (AI) collaboration. This system incorporates domain expertise, retrieval techniques, external tools and a specialized LLM fine tuned with open-forum discussions among scientists. CRISPR-GPT assists users in selecting CRISPR systems, experiment planning, designing guide RNAs, choosing delivery methods, drafting protocols, designing assays and analysing data. We showcase the potential of CRISPR-GPT by knocking out four genes with CRISPR-Cas12a in a human lung adenocarcinoma cell line and epigenetically activating two genes using CRISPR-dCas9 in a human melanoma cell line. CRISPR-GPT enables fully AI-guided gene-editing experiment design and analysis across different modalities, validating its effectiveness as an AI co-pilot in genome engineering.

  PublisherOA PDFDOI

• Semi-supervised data-integrated feature importance enhances performance and interpretability of biological classification tasks

  Bioinformatics · 2025-07-01

  articleOpen accessSenior author

  MOTIVATION: Accurate model performance on training data does not ensure alignment between the model's feature weighting patterns and human knowledge, which can limit the model's relevance and applicability. We propose Semi-Supervised Data-Integrated Feature Importance (DIFI), a method that numerically integrates a priori knowledge, represented as a sparse knowledge map, into the model's feature weighting. By incorporating the similarity between the knowledge map and the feature map into a loss function, DIFI causes the model's feature weighting to correlate with the knowledge. RESULTS: We show that DIFI can improve the performance of neural networks using two biological tasks. In the first task, cancer type prediction from gene expression profiles was guided by identities of cancer type-specific biomarkers. In the second task, enzyme/non-enzyme classification from protein sequences was guided by the locations of the catalytic residues. In both tasks, DIFI leads to improved performance and feature weighting that is interpretable. DIFI is a novel method for injecting knowledge to achieve model alignment and interpretability. AVAILABILITY AND IMPLEMENTATION: Code and models for our experiments are available at https://github.com/junwkim1/DIFI.

  PublisherOA PDFDOI

• Pool PaRTI: a PageRank-based pooling method for identifying critical residues and enhancing protein sequence representations

  Bioinformatics · 2025-05-31 · 5 citations

  articleOpen accessSenior author

  MOTIVATION: Protein language models (PLMs) produce token-level embeddings for each residue, resulting in an output matrix with dimensions that vary based on sequence length. However, downstream machine learning models typically require fixed-length input vectors, necessitating a pooling method to compress the output matrix into a single vector representation of the entire protein. Traditional pooling methods often result in substantial information loss, impacting downstream task performance. We aim to develop a pooling method that produces more expressive general-purpose protein embedding vectors while offering biological interpretability. RESULTS: We introduce Pool PaRTI, a novel pooling method that leverages internal transformer attention matrices and PageRank to assign token importance weights. Our unsupervised and parameter-free approach consistently prioritizes residues experimentally annotated as critical for function, assigning them higher importance scores. Across four diverse protein machine learning tasks, Pool PaRTI enables significant performance gains in predictive performance. Additionally, it enhances interpretability by identifying biologically relevant regions without relying on explicit structural data or annotated training. To assess generalizability, we evaluated Pool PaRTI with two encoder-only PLMs, confirming its robustness across different models. AVAILABILITY AND IMPLEMENTATION: Pool PaRTI is implemented in Python with PyTorch and is available at github.com/Helix-Research-Lab/Pool_PaRTI.git. The Pool PaRTI sequence embeddings and residue importance values for all human proteins on UniProt are available at zenodo.org/records/15036725 for ESM2 and protBERT.

  PublisherOA PDFDOI

• Innovative 3d Modeling and Well Optimization for Mauddud Formation Development in Greater Burgan Field

  2025-09-16

  article

  Abstract Traditional exploration strategies have primarily focused on conventional, well-understood geological features, i.e. simple anticlinal traps and thick reservoir formations, due to their favorable structural characteristics and economic viability. However, contemporary challenges to increase production necessitate a paradigm shift towards less conventional exploration domains, emphasizing on geologically complex and previously underutilized regions that were historically deemed economically unfeasible. Development attempts of Mauddud formation in the Greater Burgan Area has been regarded as strategically favorable. While appearing as a thin reservoir in Southeast Kuwait, one of the primary factors driving significant interest in Mauddud formation is its hydraulic connection with the underlying extensive main Burgan clastic reservoir, suggesting potential migration and mutual influence between the two formations. KOC Reservoir Study Team (RST) and Innovation &amp; Technology (I&amp;T) in partnership with SLB have been tasked with the challenge of identifying optimal locations for the new well placement, defining suitable candidates for the re-entry side-tracks, and evaluating the feasibility of repurposing old wells - originally targeted for the lower formations – for production from the Mauddud Reservoir. A critical aspect of this initiative involves the integration of all available data to confirm formation connectivity and developing a DFN model. This study presents an integrated 3D modeling approach that incorporates petrophysical, geophysical, geological, reservoir, and production data to optimize new well placement. A new fine static geological model was built for the Greater Burgan Field with a primary focus on the Mauddud formation, while the Wara and Upper Burgan formations were included at a broader scale. Several novel rock classification approaches were implemented, and image log analysis facilitated the identification of structural features and vertical heterogeneity within the formation. A supervised multivariate approach was employed to estimate the permeability model, allowing the integration of multi-domain datasets including wells logs and cores. Petrophysical rock typing, based on pore throat radius distribution, provided a detailed reservoir property characterization, improving well candidate assessment and placement optimization. Furthermore, capillary pressure-derived water saturation estimates were utilized to model the original oil in place (OOIP). At the conclusion of the comprehensive FDP study, two piloted horizontal wells were drilled and completed with multi-stage acid frac. The results of the DST were encouraging, validating the effectiveness of the revised FDP, and providing a foundation for the continued development of Mauddud formation, and paving the way for further additional wells in the future. The successful Implementation of the integrated study on the Mauddud reservoir in the Greater Burgan field, despite numerous specific uncertainties, serves as a pioneering example for future advancements in the development of challenging, geologically complex formations.

  PublisherDOI

Recent grants

• NIH Grant R01HL128170

  NIH · $3.0M · 2019

• NIH Grant R24GM061374

  NIH · $25.2M · 2020

• NIH Grant R29LM005652

  NIH · $550k · 1999

• NIH Grant U54HL117798

  NIH · $36.7M · 2018

• NIH Grant T32LM012409

  NIH · $1.5M · 2021

Frequent coauthors

• Teri E. Klein

  Stanford Medicine

  371 shared

• Scott C. Blanchard

  St. Jude Children's Research Hospital

  131 shared

• Caroline F. Thorn

  128 shared

• Michelle Whirl‐Carrillo

  Stanford University

  84 shared

• Ellen M. McDonagh

  European Bioinformatics Institute

  60 shared

• Katrin Sangkuhl

  Stanford University

  55 shared

• Howard L. McLeod

  Utah Tech University

  54 shared

• Li Gong

  Stanford University

  48 shared

Labs

• Helix GroupPI

  Not provided

Education

• Ph.D., Biochemistry

  Stanford University

  1989

• B.S., Biochemistry

  Stanford University

  1984

Awards & honors

• The Arthur Kornberg and Paul Berg Lifetime Achievement Award…
• Teaching Honor Roll, Tau Beta Pi (2020)
• Excellence in Graduate Teaching Award, Stanford Biosciences…
• Fellow, American Association for the Advancement of Science…
• Stanford Medical School Mentorship Award, Stanford Medical S…