Drug-induced cell viability prediction from LINCS-L1000 through WRFEN-XGBoost algorithm.

Area of Science:

• Computational biology
• Genomics
• Pharmacogenomics

Background:

• Predicting cancer drug response is crucial for personalized medicine.
• Traditional methods are limited by cost and sample size.
• Large gene expression datasets present opportunities for machine learning in drug sensitivity prediction.

Purpose of the Study:

• To develop a machine learning algorithm for predicting cancer cell viability and drug response.
• To leverage large-scale gene expression data for improved prediction accuracy.

Main Methods:

• Developed the WRFEN-XGBoost algorithm using LINCS-L1000 cell perturbation signatures.
• Integrated LINCS-L1000, CTRP, and Achilles datasets.
• Employed a weighted fusion algorithm for key gene selection and FEBPSO-XGBoost for cell viability prediction.

Main Results:

• The WRFEN-XGBoost model achieved a Pearson correlation of 0.83 in cell viability prediction.
• Validated drug sensitivity on NCI60 and CCLE datasets, confirming the method's effectiveness.
• Demonstrated superior performance compared to existing methods.

Conclusions:

• The developed method aids in understanding disease mechanisms and discovering novel therapies.
• This approach significantly contributes to the advancement of clinical medicine and personalized treatment strategies.