ersilia-os/pharmacogx-embeddings
Domaine:
healthcareType de record:
model
Créateur:
ers
Hôte:
Drug and gene embeddings to predict drug-gene interactions, with a focus on Africa, developed with the H3D Centre.
# PharmacoGx Embeddings
Use drug and gene embeddings to predict drug-gene interactions, with a focus on Africa.
## Project overview
This repository is related to the manuscript _AI coupled to pharmacometric modelling to tailor malaria and tuberculosis treatment in Africa_, co-led by the H3D Centre (South Africa) and the Ersilia Open Source Initiative. Please read the medRxiv preprint.
In brief, this computational work consists of two parts:
1. An ML/AI-based prioritisation of drug-pharmacogene pairs of potential relevance to African populations.
2. A pharmacometric modeling analysis of the prioritised pharmacogenes for malaria and tuberculosis drugs.
The current repository is focused on the first part, summarized in the following figure in panels A and B.
For more information:
- Check the PharmacoGx-ARSA repository for more details on the processing of 'African variants'.
- Find the app code in the PharmacoGX-App repository, or try the online demo.
- Read our comment in Nature about the launch of this project.
- To learn more about the GRADIENT projects, see this article in Drug Discovery Today.
## Knowledge graph
The pipeline starts with the compilation of a knowledge graph, and the obtention of embeddings for drugs and genes to be used as predictive features.
The knowledge graph leverages mainly PharmGKB, the Bioteque and the Chemical Checker:
* **PharmGKB** is used as a source of pharmacogenomics knowledge. We use the different concepts summarized in PharmGKB as the backbones of the knowledge graph. These include, for example, evidence levels of the variant-drug relationships, or the biogeographical groups in which a variant has been identified.
* The **Bioteque** is a knowledge graph itself integrating relations between 12 biological entities, and describing them as ML-amenable descriptors. We use it to complement the knowledge graph with additional information.
* The **Chemical Checker** is an integrative approach to obtain bioactivity descriptors fo …