FlexMS: A Unified Public Benchmark for Molecule Tandem Mass Spectrum Prediction

FlexMS is introduced as a modular public benchmark for tandem mass spectrum (MS/MS) prediction, addressing the challenges of evaluating deep learning systems in metabolomics. It standardizes the evaluation process by integrating molecular encoders, metadata conditioning, and predictor heads under a unified protocol, while also incorporating difficulty-aware diagnostics to guide model selection based on compute constraints and data variability. This benchmark is significant for practitioners as it facilitates fair comparisons of predictive tools and enhances reproducibility in algorithmic evaluations within the field.