Stanford identifies selective appetite-suppressing molecule BRP via machine learning

Researchers at Stanford University have used machine-learning models to identify BRP, a selective appetite-suppressing molecule heading towards clinical trials.

The discovery targets a persistent challenge in metabolic drug design. Existing peptide therapies for weight loss frequently produce gastrointestinal side effects. Earlier small-molecule alternatives often induced off-target toxicity or adverse central nervous system responses. The Stanford team sought a non-peptide candidate capable of modulating satiety pathways with strict pharmacological selectivity.

To search extensive chemical spaces efficiently, the researchers deployed machine-learning models trained on structural and biochemical interaction datasets. The algorithms screened millions of candidate compounds against targeted metabolic receptors, predicting binding affinities as well as secondary off-target liabilities. This predictive filtering narrowed the initial chemical library to a small subset of high-confidence candidates, leading directly to the prioritisation and synthesis of BRP.

In preclinical assays, BRP reduced caloric intake without eliciting nausea-associated behaviours or broader metabolic disruptions. Treated animal models showed measurable reductions in food consumption and body weight while maintaining lean mass. Pharmacological profiling confirmed that the molecule binds selectively to its intended metabolic target, avoiding secondary receptors tied to adverse cardiovascular and neurological responses.

The machine-learning pipeline accelerated candidate identification significantly. Rather than relying entirely on multi-year high-throughput laboratory screens, the researchers used iterative computational triage to isolate viable chemotypes in months. While computational chemistry has frequently assisted structural optimisation, BRP represents an instance where machine learning isolated an in vivo-validated metabolic lead directly from large digital libraries.

Preparations are now underway for Phase 1 clinical trials to evaluate safety, dosing, and tolerability in human subjects. As algorithms take on larger roles in metabolic drug discovery, will the selectivity profiles calculated by computational models withstand the physiological heterogeneity of clinical cohorts?