A Deep Learning Framework for the Discovery of Natural-Product Candidate Binders of Acetyl-CoA Carboxylase 2 (ACC2) with Potential Relevance to Cardiometabolic Lipid Metabolism.

Background/

Objectives

Obesity and related metabolic diseases arise from an interplay of lipid overload, insulin resistance and oxidative stress. Acetyl-CoA carboxylase 2 (ACC2) controls malonyl-CoA production and thereby gates mitochondrial fatty-acid oxidation, placing it at the intersection of lipid handling and redox-sensitive metabolic dysfunction. Dietary antioxidants such as polyphenols, flavonoids and terpenoids are increasingly studied as modulators of these pathways, yet systematic prioritization of food-derived antioxidant compounds against defined metabolic targets remains challenging. We developed an integrated deep learning and structure-based workflow to prioritize FooDB compounds with predicted ACC2-binding potential.

Methods

A curated set of 3983 ACC2 bioactivity records from ChEMBL 36 was used to train scaffold-split models, including graph neural-network and graph-Morgan fingerprint-fusion architectures. The calibrated ensemble screened 139,988 FooDB compounds; 200 candidates with predicted activity probability above 0.70 were docked against the ACC2 carboxyltransferase domain (PDB ID: 3FF6), and six prioritized complexes underwent 500 ns molecular dynamics and MM/GBSA analysis.

Results

Redocking of the co-crystallized ligand reproduced the experimental pose (RMSD 1.2 Å). Although the highest-ranked screening hits were antioxidant terpenoids and alkaloids, docking-based prioritization from the top candidates selected six larger, more polar food-derived compounds, including glycosides and two nucleotide/cofactor-like conjugates, which showed docking scores from -7.47 to -6.65 kcal/mol versus -6.21 kcal/mol for the reference ligand. Glu539 emerged as a recurrent interaction hotspot. All candidates gave more favourable MM/GBSA binding free energies than the reference (ΔG = -22.52 kcal/mol), led by FDB029596 (-35.65), FDB021568 (-34.14) and FDB017807 (-33.87 kcal/mol).

Conclusions

This workflow provides a reproducible framework for prioritizing food-derived compounds as candidate ACC2 binders relevant to obesity and metabolic disease, generating structurally supported hypotheses for biochemical and nutritional validation. The prioritized compounds are computational candidates only and require biochemical and cellular (experimental) validation before any ACC2-related biological relevance can be established.