Computer-Aided Drug Design for Ebola VP35

Problem

Identifying candidate compounds against viral proteins is slow when it depends only on wet-lab screening. Computational design can narrow the search if models are grounded in the biology of the target.

Context

A scholarly project exploring machine-learning approaches to computer-aided design against Ebola VP35. The work placed third at the KAPS 2019 hackathon in liaison with JKUAT and partners, was reported in the Standard on 18 November 2019, and progressed to a commercial outcome with San Francisco Pharmaceuticals.

Dataset

Molecular descriptors and candidate compound libraries used during the computational design exercise against the VP35 viral protein target.

Methodology

Machine-learning-assisted screening and design combining computational chemistry representations with predictive models of candidate utility against Ebola VP35.

Model / architecture

Predictive models over molecular representations for in-silico screening and ranking of candidate compounds.

Results

Third place at KAPS 2019 hackathon. The design work advanced beyond the competition to a commercial transfer with San Francisco Pharmaceuticals, demonstrating a path from hackathon research to industry uptake.

Evaluation metrics

Recognition

KAPS 2019, 3rd place

Target

Ebola VP35

Commercial outcome

San Francisco Pharmaceuticals

Challenges

Limited labelled bioactivity data, the leap from in-silico scores to experimental confirmation, and responsible communication of early-stage design work.

Impact

A demonstration that computational intelligence can participate in neglected-disease drug discovery, and a formative project in Jeff's applied research path from university hackathon to commercial pharmaceutical interest.

Tools / technology

• Python
• R
• Machine learning
• Cheminformatics