Machine Learning Predicts Lipid Lowering Potential in FDA Approved Drugs

Researchers from Southern Medical University and collaborators report the identification of FDA‑approved compounds that may lower blood lipids by combining computational screening with clinical and experimental validation. The work appears in Acta Pharmacologica Sinica and addresses the need for alternative therapies for patients who cannot use or respond to standard lipid‑lowering medications.

Screening of existing drugs using multi‑model machine learning

Starting with 3 430 FDA‑approved compounds (176 known lipid‑lowering drugs and 3 254 others) the team extracted molecular descriptors and fingerprints from SMILES codes and physicochemical data. Features were narrowed via Spearman correlation and LASSO regression. A suite of 68 machine learning models including random forest, support vector machine, gradient boosting and elastic net combinations were trained and evaluated using AUC accuracy F1 recall specificity metrics. Top performing models reached AUC ≈ 0.886 and accuracy ≈ 0.888. Drugs predicted positive in at least 8 of the top 10 models were flagged, yielding 29 repurposing candidates.

Retrospective clinical data supports four candidates

Medical records from nearly 25 years (June 1998–May 2024) at Zhujiang Hospital were reviewed for patients treated with model‑predicted drugs. Four agents – Argatroban, Levoxyl (levothyroxine), oseltamivir and thiamine – showed statistically significant lipid changes. For Argatroban (n = 63) LDL fell by 33 percent (2.96 mmol/L to 1.98 mmol/L), total cholesterol by 25 percent (4.68 to 3.51 mmol/L), and triglycerides also declined (all P < 1 × 10⁻⁸). Levoxyl users (n = 87) had LDL and TC reductions of 16 percent and 12 percent respectively. Oseltamivir and thiamine exhibited moderate but significant lipid effects.

Mouse studies confirm lipid modulation for multiple drugs

Sixteen drugs selected from computational predictions and clinical evidence were tested in male C57BL/6 mice. Argatroban and Promega reduced total cholesterol by ~10 percent. Levoxyl and sulfaphenazole each lowered triglycerides by ~27–29 percent. Prasterone, alpha tocopherol acetate, sorafenib, Cedazuridine and Promega significantly increased HDL levels—Prasterone produced the largest rise (~24 percent). Unexpectedly LDL increased modestly in some drug groups (e.g. procarbazine, dimenhydrinate, Promega) compared to control. All findings are based on animal preclinical data and require further investigation.

Molecular docking and dynamics highlight possible mechanisms

Seven promising drugs (Argatroban, Promega, sulfaphenazole, sorafenib, prasterone, Levoxyl, alpha‑tocopherol acetate) were docked against 12 lipid metabolism targets including HMG‑CoA reductase, coagulation factor X, serotonin receptors (HTR2A/C, 5‑HT4R), thyroid hormone receptors (TRα, TRβ), MTP, RXRα and COX‑2. Argatroban bound tightly to coagulation factor X (≈ –7.6 kcal/mol) forming hydrophobic interactions and stable hydrogen bonds in molecular dynamics simulations. Levoxyl showed high affinity for TRα; sulfaphenazole bound serotonin receptor subtypes; prasterone engaged RXRα and COX‑2; Promega associated with MTP; sorafenib showed affinity to HMG‑CoA reductase. Results suggest multiple candidate drugs act via distinct lipid pathways.

Study strengths limitations and future directions

This study integrates large‑scale predictive modelling with clinical and animal data and mechanistic docking to identify 29 candidate lipid‑lowering drugs. Argatroban, Levoxyl and sulfaphenazole emerge as particularly promising based on consistency across methods. Limitations include reliance on retrospective single‑centre clinical data and murine models that may not translate to humans. Randomized controlled trials and biochemical studies are needed to confirm efficacy and safety in humans. The heterogeneity of positive control drugs suggests stratified predictive models may increase future precision. The framework may be applied to drug repurposing in other therapeutic areas.

Reference: Chen J‑h, Li K‑x, Fan C‑f, et al. Integration of machine learning and experimental validation reveals new lipid‑lowering drug candidates. Acta Pharmacol Sin. 2025. doi:10.1038/s41401‑025‑01539‑1

This content includes text that has been generated with the assistance of AI. Technology Networks' AI policy can be found here.