Predicting drug-perturbed transcriptional responses using multi-conditional diffusion transformer.

Area of Science:

• Computational Biology
• Genomics
• Artificial Intelligence

Background:

• Drug-perturbed transcriptomes are crucial for personalized medicine and drug discovery.
• Current high-throughput screening methods are expensive and time-consuming.

Purpose of the Study:

• To introduce a novel multi-condition diffusion transformer model, PertDiT, for generating perturbed transcriptomes.
• To enable conditional generation of transcriptomes based on drug text information.

Main Methods:

• Developed PertDiT, integrating transformer architecture with large language models for text representation.
• Incorporated novel perturbation and transcriptome fusion modules.
• Designed CrossDiT and CatCrossDiT network structures for drug discovery and personalized medicine.

Main Results:

• PertDiT demonstrated superior performance in post-perturbation transcriptome reconstruction.
• The model accurately predicted perturbation-induced transcriptional changes.
• Outperformed existing methods across comprehensive metrics and data splitting strategies.

Conclusions:

• PertDiT offers an efficient and effective approach for generating drug-perturbed transcriptomes.
• The model shows significant promise for advancing drug discovery and personalized medicine.
• The model's architecture and effectiveness were rigorously validated.