Format

Send to

Choose Destination
Cell Chem Biol. 2019 Apr 8. pii: S2451-9456(19)30107-2. doi: 10.1016/j.chembiol.2019.03.011. [Epub ahead of print]

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

Author information

1
Institute for Molecular Medicine Finland (FIMM), HiLIFE, University of Helsinki, 00290 Helsinki, Finland.
2
Institute for Molecular Medicine Finland (FIMM), HiLIFE, University of Helsinki, 00290 Helsinki, Finland; Department of Mathematics and Statistics, University of Turku, 20500 Turku, Finland.
3
The Miami Project to Cure Paralysis, Peggy and Harold Katz Family Drug Discovery Center, Sylvester Comprehensive Cancer Center, and Departments of Neurological Surgery and Medicine, University of Miami Miller School of Medicine, Miami, FL 33136, USA; Truvitech LLC, Miami, FL 33136, USA. Electronic address: halali@med.miami.edu.
4
Institute for Molecular Medicine Finland (FIMM), HiLIFE, University of Helsinki, 00290 Helsinki, Finland; Biotech Research & Innovation Centre (BRIC) and Novo Nordisk Foundation Center for Stem Cell Biology (DanStem), University of Copenhagen, 2200 Copenhagen N, Denmark. Electronic address: krister.wennerberg@bric.ku.dk.

Abstract

The lack of functional understanding of most mutations in cancer, combined with the non-druggability of most proteins, challenge genomics-based identification of oncology drug targets. We implemented a machine-learning-based approach (idTRAX), which relates cell-based screening of small-molecule compounds to their kinase inhibition data, to directly identify effective and readily druggable targets. We applied idTRAX to triple-negative breast cancer cell lines and efficiently identified cancer-selective targets. For example, we found that inhibiting AKT selectively kills MFM-223 and CAL148 cells, while inhibiting FGFR2 only kills MFM-223. Since the effects of catalytically inhibiting a protein can diverge from those of reducing its levels, targets identified by idTRAX frequently differ from those identified through gene knockout/knockdown methods. This is critical if the purpose is to identify targets specifically for small-molecule drug development, whereby idTRAX may produce fewer false-positives. The rapid nature of the approach suggests that it may be applicable in personalizing therapy.

KEYWORDS:

AI; AURKA; Akt; FGFR; PKIS; TNBC; cancer cell line; dependency; drug screening; gene silencing; kinase; kinase inhibitors; machine learning; target deconvolution

Supplemental Content

Full text links

Icon for Elsevier Science
Loading ...
Support Center