Supervised learning with word embeddings derived from PubMed captures latent knowledge about protein kinases and...

Area of Science:

• Computational biology
• Bioinformatics
• Oncology

Background:

• Protein kinases (PKs) are crucial targets in cancer therapy, with many inhibitors in clinical trials.
• Identifying novel PKIs for specific cancers remains a significant challenge in drug discovery.

Purpose of the Study:

• To develop and validate a machine learning approach for predicting efficacious protein kinase inhibitors (PKIs) for various cancers.
• To leverage natural language processing (NLP) and clinical trial data to identify novel therapeutic targets.

Main Methods:

• Representing PKs and cancers as 100-dimensional semantic vectors derived from PubMed abstracts.
• Utilizing phase I-IV clinical trial data from ClinicalTrials.gov to build a random forest classification model.
• Training and validating the model on historical data to assess prediction accuracy.

Main Results:

• The developed approach accurately predicts associations between PKs and specific cancers.
• Predictions of PK-cancer relationships were achieved with high accuracy, years in advance.
• The model demonstrates the potential for identifying novel PKIs for targeted cancer therapy.

Conclusions:

• This NLP and machine learning tool can predict the relevance of inhibiting specific PKs for cancer treatment.
• The approach supports the design of focused clinical trials for discovering new PKIs.
• This method offers a promising strategy for advancing precision oncology and drug development.