PKM2Pred: An AI Tool for Rapid Identification and Potency Estimation of PKM2-Targeting Anticancer Compounds
Aryan Raj Saxena1, Palak Singla1, Arya Chakraborty2
1Advanced BioComputing Lab, Department of Bioengineering and Biotechnology, Birla Institute of Technology Mesra, Ranchi, 835215 Jharkhand, India.
Abstract:
Cancer continues to pose a major global health challenge due to its metabolic complexity. Pyruvate Kinase M2 (PKM2), a key glycolytic enzyme, is central to tumor progression and metastasis. To facilitate targeted drug discovery, we introduce PKM2Pred (https://pkm2pred.vercel.app/), a machine learning based freely accessible web server that classifies compounds as activators, inhibitors, or decoys and predicts their AC50 range. Built on a Random Forest classifier, the model achieved 94% accuracy and a Matthews Correlation Coefficient of 90.02%. A bootstrapped regression model estimated bioactivity ranges with confidence intervals, offering flexibility between prediction and range. The top three key molecular descriptors, such as WTPT-5, SRW9, and nHeteroRing, emerged as the most important statistical descriptors based on their percentage importance of 12.5, 8.2, and 5.8, respectively. Thus, PKM2Pred offers rapid, reliable, and cost-effective computational insight for anticancer drug discovery.
Insights
A new machine learning tool, PKM2Pred, aids anticancer drug discovery by classifying compounds and predicting their bioactivity. This freely accessible web server offers rapid computational insights into Pyruvate Kinase M2 (PKM2) modulation.
Area of Science:
- Computational chemistry and bioinformatics
- Drug discovery and development
- Cancer metabolism research
Background:
- Cancer's metabolic complexity presents a significant challenge for drug development.
- Pyruvate Kinase M2 (PKM2), a key glycolytic enzyme, plays a crucial role in tumor progression and metastasis.
- Targeted drug discovery requires efficient methods to identify modulators of key cancer-related proteins like PKM2.
Purpose of the Study:
- To introduce PKM2Pred, a freely accessible web server for computational drug discovery targeting PKM2.
- To classify chemical compounds as PKM2 activators, inhibitors, or decoys.
- To predict the AC50 bioactivity range of compounds with confidence intervals.
Main Methods:
- Development of a machine learning model based on the Random Forest classifier.
- Utilized key molecular descriptors (WTPT-5, SRW9, nHeteroRing) for prediction.
- Employed a bootstrapped regression model to estimate bioactivity ranges and confidence intervals.
Main Results:
- The PKM2Pred model achieved 94% accuracy and a Matthews Correlation Coefficient of 90.02%.
- Identified WTPT-5, SRW9, and nHeteroRing as the most important molecular descriptors.
- The server provides rapid, reliable, and cost-effective computational insights for drug discovery.
Conclusions:
- PKM2Pred offers a valuable computational tool for accelerating anticancer drug discovery.
- The web server facilitates efficient screening and prediction of PKM2-targeting compounds.
- This approach provides significant cost and time savings in the drug development pipeline.
