Skip to main content

Welcome to nSIDES

nSIDES is the home for the drug side effect and drug interaction resources made available from the Tatonetti Lab. Below you will find descriptions of each of the resources with links to download the data and access the code. Please reach out if you have any questions and share how you're using these resources! You can reach us via email or Twitter.


Drug side effects extracted from structured product labels.

OnSIDES is a database of adverse drug events extracted from drug labels created by fine-tuning a PubMedBERT language model on 200 manually curated labels available from Denmer-Fushman et al.. This comprehensive database will be updated quarterly, and currently contains more than 3.6 million drug-ADE pairs for 2,793 drug ingredients extracted from 46,686 labels, processed from all of the labels available to download from DailyMed as of November 2023. Additionally, we now provide a number of complementary databases constructed using a similar method - OnSIDES-INTL, adverse drug events extracted from drug labels of other nations/regions (Japan, UK, EU), and OnSIDES-PED, adverse drug events specifically noted for pediatric patients in drug labels. We have recently released a preprint on medRxiv with a full description of the data, methods and analyses.

Model Accuracy

Our fine-tuned language model achieves an F1 score of 0.90, AUROC of 0.92, and AUPR of 0.95 at extracting effects from the ADVERSE REACTIONS section of the FDA drug label. For the BOXED WARNINGS sections, the model achieves a F1 score of 0.71, AUROC of 0.85, and AUPR of 0.72. Compared against the reference standard using the official evaluation script for TAC 2017, the model achieves a Micro-F1 score of 0.87 and a Macro-F1 of 0.85.

Performance metrics evaluated against the TAC gold standard

MetricTAC (Best Model)SIDER 4.1OnSIDES v1.0.0OnSIDES v2/3.0.0
F1 Score82.1974.3682.0187.54

Roberts, Demner-Fushman, & Tonning, Overview of the TAC 2017


The latest database versions are available as a flat files in CSV format. Previous database versions can be accessed under Releases. A DDL (load_onsides_db.sql) is provided to load the CSV files into a SQL schema.

Source Code

Release notes, validation study results, and code for generating the models and data for Onsides is available on the project's GitHub.


A resource of pediatric drug safety signals.

Adverse drug events are responsible for up to 10% of pedaitric hospitalizations and drug side effects are primaryYeah safety concerns in pediatrics. Unfortunately, due to the structural and ethical challenges of including children in clinical trials, there is little information available. We invented a method that leverages post-marketing adverse drug event reports to identify drug safety signals that vary across the developmental phases of childhood. We systematically applied this method across all adverse events and all childhood developmental phase to produce a database of pediatric drug safety signals, called KidSIDES. You can access the manuscript, source code, data files, and web portal for KidSIDES below.



The PDSPortal is a RShiny application we created to enable browsing of the pediatric drug safety signals. You can access it at


MySQL and SQLite Files

Flat Files

Source Code

The source code for the published study is available in pediatric_ade_database_study. Code for the PDSPortal and the PDSPortal_case_studies are also provided.


A resource of drug safety signals disproportionately affecting women.

Adverse drug reactions (ADRs) are the fourth leading cause of death in the US. Although women take longer to metabolize medications and experience twice the risk of developing ADRs compared to men, these sex differences are not comprehensively understood. Real-world clinical data provides an opportunity to estimate safety effects in otherwise understudied populations, ie. women. These data, however, are subject to confounding biases and correlated covariates. We present AwareDX, a pharmacovigilance algorithm that leverages advances in machine learning to study sex risks. Our algorithm mitigates these biases and quantifies the differential risk of a drug causing an adverse event in either men or women. We present a resource of 20,817 adverse drug effects posing sex specific risks. We independently validated our algorithm against known pharmacogenetic mechanisms of genes that are sex-differentially expressed. AwareDX presents an opportunity to minimize adverse events by tailoring drug prescription and dosage to sex.

Data and Source

The data is available here and source code is available from the sex_risks github repository.


Chandak P, Tatonetti NP. Using Machine Learning to Identify Adverse Drug Effects Posing Increased Risk to Women. Patterns (N Y). 2020 Oct 9;1(7):100108. doi: 10.1016/j.patter.2020.100108. Epub 2020 Sep 22. PMID: 33179017; PMCID: PMC7654817.


Off label drug side effect and drug-drug interaction safety signals.

Drug side effects and drug-drug interactions were mined from publicly available data. OffSIDES (OFF label drug SIDE effectS) is a database of drug side-effects that were found, but are not listed on the official FDA label. TwoSIDES is the only comprehensive database drug-drug-effect relationships. Over 3,300 drugs and 63,000 combinations connected to millions of potential adverse reactions.

Note that these resources are quite a bit out of date. We will be updating them for 2022 and then including them our routine quarterly updates going forward. As the repositories and project pages are built out for these projects, they will be posted here.


The data are available as flat files in this directory.


Side effect signals for combinations of drugs (3+).

ManySIDES (currently called nsides v0.1 but in the process of transitioning to its new name) is our resource for drug combination side effects and is currently under active development. You can read more about v0.1 in the release notes and the code repository.