Russ Biagio Altman

The paper shows homology-based variant-effect predictors fail for CYP enzymes and proposes an improved model that still lacks clinical relevance.

SPHERE enables the use of sensitive datasets in AI and open science by creating synthetic twins that preserve statistical properties and privacy.

Inverse FoldDir designs protein sequences compatible with a given backbone structure, achieving high accuracy and experimental validation in nanobody redesign.

AutoScreen improves gene prioritization in functional genomics by integrating AI agents for pre-screen design and post-screen analysis, achieving higher hit recovery and precision than existing methods.

The paper identifies high attention sites in protein language models that predict functional and pathogenic regions in proteins.

GATSBI improves protein embeddings by using biologically relevant data splits, showing better performance for understudied proteins in various tasks.

This paper introduces PIGLET, a graph transformer method for drug-target interaction prediction that outperforms existing models on a drug-based split of the Human dataset.

The paper introduces a dataset of predicted binding pockets across the human proteome and a machine learning model to score and filter these pockets, outperforming existing methods.

This paper identifies High Attention (HA) sites in protein sequences using the ESM model to improve protein family classification and function prediction.

This paper introduces PARSE, a method for protein functional site annotation using local structure embeddings, achieving high accuracy in predicting catalytic residues and novel functions.

This study evaluates LLMs' ability to infer biochemical reactions and disease associations from protein pathways, finding poor performance and introducing a benchmark dataset.

CRISPR-GPT automates CRISPR-based gene-editing design and analysis by integrating LLMs with domain expertise, demonstrating effectiveness in gene knockout and activation experiments.

The paper uses self-supervised deep learning to cluster local 3D protein structures, creating a motif lexicon that improves structure prediction, drug off-target prediction, and variant analysis.

Pool PaRTI improves protein sequence representations by using PageRank on transformer attention matrices to identify critical residues, enhancing performance and interpretability across ML tasks.

The paper uses EHRs and biobanks to identify pharmacogenetic associations with drug response in cardiometabolic diseases, finding novel genetic predictors for personalized treatment.

This paper introduces two methods for GWAS using disease liabilities from deep neural networks, identifying more loci and improving prediction of newly diagnosed cases.

This study identifies E2F4 as a key regulator of colon cancer resistance to irinotecan by showing its role in drug resistance through shRNA knockdown in PDX models.

SubCell is a deep learning model for fluorescence microscopy that captures cellular morphology and protein organization, outperforming existing methods and enabling a multimodal approach to gene function.

The paper uses network-based representation learning to identify risk genes and functional pathways in primary open-angle glaucoma using protein-protein interaction networks and gene embeddings.

The paper introduces a phenotype-genotype reference map for replicating GWAS results in biobank data and demonstrates its utility in detecting data corruption and assessing reproducibility.

COLLAPSE learns deep representations of protein structural sites using 3D atomic positions and evolutionary signals, achieving state-of-the-art performance in function prediction and site annotation.

The paper constructs disease-specific cytokine profiles by linking disease genes to immune responses using protein-protein interaction networks and association scores.

This study evaluates how training data composition affects protein structure analysis algorithms across three tasks, finding that including diverse structure types improves performance.

The paper analyzes smoking and sex effects on gene expression across multiple tissues using public data, identifying tissue-specific and sex-related gene expression changes.

The paper uses CRISPRa screening and real-world data to identify KCNA6 as a neuronal entry factor for SARS-CoV-2 and shows dalfampridine can suppress viral entry.

CoronaCentral organizes and analyzes SARS-CoV-2 literature using machine learning to categorize and track research trends, media attention, and article types.

The paper infers sex labels from gene expression data across human and mouse samples and identifies sex bias in drug studies.

This paper quantifies adverse drug reaction severity using social media data and presents scores correlated with real-world outcomes.

The paper presents a user-friendly web tool for drug candidate discovery through a biomedical knowledge graph and visualization, showing it can identify approved repurposed drugs.

The paper analyzes pharmacogenetics in the UK Biobank, finding that most individuals may have atypical drug responses and non-European populations have more deleterious variants.