ProtoQSAR and MolDrug, together with IATA-CSIC, develop a computational tool to accelerate the search for new drugs against type 2 diabetes
DPPPRED-IV, a machine learning-based system that enables the rapid and reliable identification of new active compounds against the DPP4 enzyme, a key target in the treatment of type 2 diabetes.
Type 2 diabetes mellitus (T2DM) is a chronic, multifactorial disease affecting more than 400 million people worldwide. Although effective therapies exist, such as dipeptidyl peptidase 4 (DPP4) enzyme inhibitors, the development of new drugs remains a challenge due to the vastness of chemical space and the high cost of experimental trials. In this context, a research team led by associated biotech companies ProtoQSAR and MolDrug AI Systems, in collaboration with the group led by Dr. Yolanda Sanz at the Institute of Agrochemistry and Food Technology of the CSIC (IATA-CSIC), has developed a new computational tool that promises to transform the early-stage screening of therapeutic candidates against type 2 diabetes.
The tool, named DPPPRED-IV, has been published in the international scientific journal International Journal of Molecular Sciences and is already freely available through the ChemoPredictionSuite platform. It is a prediction system based on QSAR (Quantitative Structure-Activity Relationship) models, combining binary classification and regression methods to estimate both the probability that a compound acts as a DPP4 inhibitor and its potency (IC₅₀).
A combination of artificial intelligence and experimental data
To build DPPPRED-IV, the researchers compiled and curated a dataset of more than 4,000 bioactive compounds sourced from the ChEMBL database, one of the most recognized databases internationally. Based on this data, genetic algorithms were applied to select the most relevant molecular descriptors, and several machine learning models were trained and combined into a weighted voting system to improve prediction accuracy.
It is also worth noting that the models were validated using real experimental data. Following a virtual screening of a commercial database, 29 compounds with varying probabilities of activity against this enzyme were selected. The results confirmed that the tool achieves a prediction accuracy of around 70%, demonstrating its usefulness in reducing the number of trials required and directing research toward more promising compounds.
An open tool at the service of research
DPPPRED-IV is now publicly accessible within the ChemoPredictionSuite platform, allowing any user to input chemical structures and obtain immediate predictions on their inhibitory potential against the DPP4 enzyme. It also provides information on the model’s applicability domain, helping to interpret the results obtained with greater rigor.
This advance represents a significant step toward more efficient, ethical and sustainable research, by helping to reduce both the number of compounds requiring synthesis and the use of animal models in the early stages of pharmaceutical development.