Nicholas P. Tatonetti

PESTO classifies gene-phenotype associations as Novel, Hypothesized, Existing, or Established using a large language model and structured genetic evidence, identifying 7% as Novel and 21% as Hypothesized in 940 associations.

MaxGCP improves GWAS power by optimizing phenotype definitions to maximize coheritability with complex traits in observational data.

BioMADE predicts Torsades de Pointes risk using biologically informed molecular representations, outperforming chemical-only models.

WebGWAS provides a web tool to quickly generate approximate GWAS summary statistics for custom phenotypes without requiring individual-level data.

The paper identifies cell-type specific single-cell signatures that reveal nephrotoxic drug effects and improves prediction of drug-induced kidney injury compared to bulk RNA sequencing.

This study identifies CFTR-related defects in idiopathic pancreatitis and shows that ivacaftor can reverse these defects in patient-derived models.

This paper introduces Indirect GWAS, a method that accelerates biobank-scale GWAS by computing statistics for custom phenotypes using summary statistics, reducing runtime significantly.

This study shows that Norrin/β-catenin signaling in mural cells increases Lama2 expression, which promotes neurovascular unit assembly and barrier maturation.

This paper introduces TissueTox, an algorithm that predicts tissue toxicity of druggable proteins using multi-omic features and distinguishes between successful and failed drugs.

This study uses rapid diagnostics and metatranscriptomics to analyze SARS-CoV-2 infection, revealing viral evolution, host responses, and drug interactions in NYC samples.

This study presents a hierarchical machine learning model to predict drug-drug interaction-associated adverse drug reactions and provides pharmacological insights, achieving ~91% accuracy.

The paper compares machine learning algorithms for identifying acute ischemic stroke patients using EHR data, finding high accuracy with logistic regression and demonstrating generalizability.

This study shows that transfer learning from synthetic OCT image simulations improves classification of esophageal tissue in BE patients compared to models trained on clinical data with uncertain labels.

EdiTyper analyzes targeted sequencing data from genome editing experiments to detect mutations and indels with high accuracy and scalability.

The paper improves GWAS power for stroke by using machine learning to assign disease probabilities from electronic health records, discovering a novel variant associated with intracerebral hemorrhage.

This paper predicts genetic ancestry for 2.6 million NYC patients using only clinical data from EHRs, achieving over 80% accuracy for unknown or other ancestries.

The paper updates an algorithm to extract familial relationships from EHRs, increasing the number of relationships and improving algorithm speed and privacy.

The paper introduces a Python package called cohorts for managing clinical and biomarker data to improve standardization and reproducibility in precision medicine studies.

This paper introduces SIDERctrl, a method to predict negative control drugs for side effects, reducing false negatives by one-third using AEOLUS data.

RecNW is a fast exact aligner for targeted sequencing that outperforms existing tools by using recurrence and unique read alignment.

VenomKB v2.0 is an open-access resource for venom bioactivities, sequences, and classifications, providing tools for computational toxinology and drug discovery.

The paper uses next-of-kin data from electronic health records to estimate heritability for 500 disease phenotypes, finding results consistent with existing literature.