A phenome-guided drug repositioning through a latent variable model.

Area of Science:

• Pharmacogenomics and computational biology.
• Utilizing large-scale data for drug discovery.

Background:

• The human phenome, encompassing all phenotypes, offers insights into disease correlations and drug repositioning.
• Existing research explores phenome-genome relationships and drug similarity for therapeutic discovery.
• This study posits that interconnected phenotypes within the phenome can reveal new drug indications.

Purpose of the Study:

• To develop a generative model for comprehensive phenome analysis.
• To identify novel therapeutic indications for existing drugs through probabilistic phenome distribution.
• To leverage Latent Dirichlet Allocation (LDA) for exploring new drug uses.

Main Methods:

• Developed and optimized a Latent Dirichlet Allocation (LDA) model using phenome data from the Side Effect Resource (SIDER).
• Validated model performance by assessing its recovery potential on a perturbed drug-phenotype matrix.
• Applied the optimized LDA model to the entire phenome to identify drug repositioning candidates and suggest alternative indications.

Main Results:

• The LDA model achieved a 70% recovery rate for probabilistically significant drug-phenotype pairs.
• Successfully identified approved indications for 6 drugs not previously listed in SIDER.
• Suggested alternative treatment options for 908 drugs, with several findings supported by scientific literature.

Conclusions:

• The phenome can be effectively analyzed using generative models to uncover probabilistic drug-use associations.
• Latent Dirichlet Allocation (LDA) functions as an enrichment tool, narrowing the search space for novel drug indications.
• This approach facilitates the exploration of new therapeutic applications for existing medications.