Hybrid semantic recommender system for chemical compounds in large-scale datasets

Area of Science:

• Cheminformatics
• Computational Chemistry
• Data Science

Background:

• The growing number of chemical compounds presents a significant challenge for researchers seeking to identify relevant substances.
• Efficiently exploring large chemical datasets is crucial for scientific discovery and innovation.

Purpose of the Study:

• To develop and evaluate a novel recommender system for identifying chemical compounds of interest to researchers.
• To address the challenges posed by large-scale chemical data exploration.

Main Methods:

• Implementation of a hybrid recommender model integrating collaborative filtering (Alternating Least Squares, Bayesian Personalized Ranking) with a content-based approach.
• Utilizing semantic similarity from the ChEBI ontology for the content-based component.
• Assessment on the CheRM-20 implicit dataset comprising over 16,000 chemical compounds.

Main Results:

• The hybrid recommender system demonstrated superior performance compared to collaborative-filtering algorithms alone.
• Improvements exceeding ten percentage points were observed across multiple evaluation metrics.
• The model effectively retrieves a ranked list of relevant chemical compounds.

Conclusions:

• Hybrid recommender systems offer a powerful solution for navigating and extracting value from extensive chemical compound datasets.
• The proposed model enhances the efficiency and accuracy of chemical compound discovery for scientific research.
• Semantic similarity integration proves beneficial for improving recommender system performance in cheminformatics.