Phenotypic Screening Combined with Machine Learning for Efficient Identification of Breast Cancer-Selective...

Area of Science:

• Oncology
• Genomics
• Pharmacology
• Bioinformatics

Background:

• Genomics-based cancer drug target identification faces challenges due to limited functional understanding of mutations and protein druggability.
• Most proteins are not readily druggable, hindering the development of targeted cancer therapies.

Purpose of the Study:

• To implement and validate a machine-learning approach (idTRAX) for directly identifying effective and druggable oncology targets.
• To compare idTRAX-identified targets with those found through gene knockout/knockdown methods for small-molecule drug development.

Main Methods:

• Developed and applied idTRAX, a machine-learning model correlating cell-based small-molecule screening with kinase inhibition data.
• Utilized idTRAX on triple-negative breast cancer cell lines to identify cancer-selective targets.

Main Results:

• idTRAX efficiently identified cancer-selective targets, such as AKT and FGFR2, in triple-negative breast cancer cell lines.
• Inhibition of AKT selectively killed MFM-223 and CAL148 cells, while FGFR2 inhibition only affected MFM-223 cells.
• Targets identified by idTRAX often differ from gene knockout/knockdown methods, potentially reducing false positives for small-molecule drug development.

Conclusions:

• idTRAX offers a direct and efficient method for identifying druggable oncology targets, particularly for small-molecule drug development.
• The approach's speed suggests potential applications in personalized cancer therapy.
• idTRAX may yield fewer false positives compared to traditional gene-based methods when seeking targets for small-molecule inhibitors.