Prediction and Prioritisation of Novel Anthelmintic Candidates from Public Databases Using Deep Learning and...

Area of Science:

• Veterinary Parasitology
• Computational Chemistry
• Drug Discovery

Background:

• Parasitic nematodes cause significant animal health and economic losses.
• Widespread anthelmintic drug resistance severely limits current treatment options.
• Novel anthelmintic compounds with unique mechanisms of action are urgently needed.

Purpose of the Study:

• To evaluate an in silico (computational) approach for accelerating the discovery of new anthelmintics.
• To utilize machine learning for identifying potential drug candidates against the parasitic nematode *Haemonchus contortus*.

Main Methods:

• A supervised machine learning model (multi-layer perceptron classifier) was trained on 15,000 small-molecule compounds with known bioactivity against *H. contortus*.
• The model was validated using high-throughput screening and literature data, achieving 83% precision and 81% recall for active compounds.
• The trained model screened 14.2 million compounds from the ZINC15 database to predict nematocidal candidates.

Main Results:

• Machine learning model demonstrated high performance in identifying active compounds despite imbalanced training data.
• In silico screening identified promising candidates for novel anthelmintics.
• Experimental validation of 10 candidates showed significant inhibition of *H. contortus* larvae and adults in vitro.
• Two compounds exhibited high potency, warranting further investigation as lead anthelmintic candidates.

Conclusions:

• Machine learning-based in silico approaches can significantly accelerate the prediction and prioritization of anthelmintic small molecules.
• This computational strategy aids in the discovery of novel compounds to combat drug-resistant parasitic nematodes.
• The identified lead compounds offer potential for developing new strategies against animal parasitic infections.