MolGene-E: Inverse Molecular Design to Modulate Single Cell Transcriptomics

Area of Science:

• Computational Biology
• Pharmacology
• Machine Learning

Background:

• Systems pharmacology aims to restore diseased cells to healthy states, addressing unmet medical needs beyond conventional drug discovery.
• Single-cell transcriptomics offers detailed cellular state mapping but presents challenges due to data noise, heterogeneity, scarcity, and high dimensionality.
• Current machine learning methods are insufficient for designing drug molecules using single-cell omics data.

Purpose of the Study:

• To develop a novel deep generative framework, MolGene-E, capable of designing new drug molecules from single-cell transcriptomics data.
• To address the limitations of existing machine learning methods in handling noisy, high-dimensional single-cell omics data for drug discovery.

Main Methods:

• Developed MolGene-E, a deep generative framework integrating two novel models: a cross-modal model for harmonizing and denoising transcriptomics data, and a contrastive learning-based generative model for molecule design.
• Utilized chemical-perturbed bulk and single-cell transcriptomics data as input for the generative framework.
• Validated generated molecules using CRISPR target knock-out experiments.

Main Results:

• MolGene-E effectively harmonizes and denoises complex single-cell transcriptomics data.
• The framework generates high-quality, hit-like molecules based on gene expression profiles.
• MolGene-E demonstrated state-of-the-art performance in zero-shot molecular generation, outperforming baseline methods across diverse metrics.

Conclusions:

• MolGene-E represents a significant advancement in applying machine learning to single-cell omics data for drug discovery.
• The framework shows potential as a powerful new tool for identifying novel drug candidates, particularly for complex diseases.
• This approach overcomes key challenges associated with single-cell data, paving the way for more effective drug design.