Machine Learning Models for the Prediction of Chemotherapy-Induced Peripheral Neuropathy

Abstract

Purpose

Chemotherapy-induced peripheral neuropathy (CIPN) is a common adverse side effect of cancer chemotherapy that can be life debilitating and cause extreme pain. The multifactorial and poorly understood mechanisms of toxicity have impeded the identification of novel treatment strategies. Computational models of drug neurotoxicity could be implemented in early drug discovery to screen for high-risk compounds and select safer drug candidates for further development.

Methods

Quantitative-structure toxicity relationship (QSTR) models were developed to predict the incidence of PN. A manually curated library of 95 approved drugs were used to develop the model. Molecular descriptors sensitive to the incidence of PN were identified to provide insights into structural modifications to reduce neurotoxicity. The incidence of PN was predicted for 60 antineoplastic drug candidates currently under clinical investigation.

Results

The number of aromatic nitrogens was identified as the most important molecular descriptor. The chemical transformation of aromatic nitrogens to carbons reduced the predicted PN incidence of bortezomib from 32.3% to 21.1%. Antineoplastic drug candidates were categorized into three groups (high, medium, low) based on their predicted PN incidence.

Conclusions

QSTR models were developed to link physicochemical descriptors of compounds with PN incidence, which can be utilized during drug candidate selection to reduce neurotoxicity.