Genomics in Drug Discovery

Omics data

Use of machine learning techniques

Author

Chi Zhang

Published

January 3, 2024

Why do we need precision medicine? Late-stage failures cost the most, and small improvements in failure rates at early stage yield largest savings - use better targets.

As of 2016, 10-15% targets have genetic data; increased to 50%, expect 13-15% cost reduction.

Not all genes are targets, and not all targets are genes.

FTO (fat mass and obesity gene). Can search FTO in clinical trial gov website.

Overall nearly 60% are pursued by multiple companies, 26% by more than 5 companies: once a drug is made a target, redundancy is high. But before a target is proven, high diversity and novelty.

Machine learning

Genomics data alone are insufficient for therapeutic development. How they interact with other types of data such as compounds, proteins, EHR, images, texts etc need to be investigated.

Target discovery: identify the molecure that can be targeted by a drug to produce a therapeutic effect, such as inhibition, to block the disease process.

Therapeutic discovery: design potent therapeutic agents to modulate the target and block disease pathway. ML can be used to predict drug response in cell lines. Drug combination screening.

During clinical studies, ML can help characterize patient groups and identify eligible patients from gene expression data and EHRs.

During post-market studies, mining EHR and other RWD to provid additional evidence, such as patients’ drug response given different patient characteristics.

Supervised learning

Regression and classification, e.g. 

  • drug sensitivity prediction
  • gene expression signitures that predict clinical trial success

Unsupervised learning

Clustering, e.g. 

  • feature reduction in single-cell data to identify cell types
  • cell types and biomarkers from single-cell RNA data

Example: drug sensitivity predictive model. Identify biomarkers and build drug sensitivity predictive models using preclinical data, then apply to patients in early-stage clinical trials. Once validated, the model can be used for patient stratification and disease indicaation selection to support clinical development of a drug.

(Example from (Vamathevan et al. 2019))

References

Vamathevan, Jessica, Dominic Clark, Paul Czodrowski, Ian Dunham, Edgardo Ferran, George Lee, Bin Li, et al. 2019. Applications of machine learning in drug discovery and development.” Nature Reviews of Drug Discovery 18 (6): 463–77. https://doi.org/10.1038/s41573-019-0024-5.

https://www.iqvia.com/-/media/iqvia/pdfs/library/white-papers/machine-learning-applications-for-therapeutic-tasks.pdf