MillerBind: A Harmonic Framework for Protein-Ligand Binding Affinity Prediction with State-of-the-Art Performance on CASF-2016

Accurate prediction of protein-ligand binding affinity remains a central challenge in computational drug discovery. We present MillerBind, a novel scoring function based on a proprietary mathematical framework-Miller Harmonic Mathematics (MHM)-that encodes atomic and amino acid interactions using periodic harmonic classification and compatibility scoring. Evaluated on the CASF-2016 benchmark (285 held-out complexes), MillerBind achieves a Pearson correlation of R = 0.890 and mean absolute error of MAE = 0.78 pKd units, surpassing established methods including RF-Score (R = 0.80), OnionNet-2 (R = 0.82), PIGNet (R = 0.83), IGN (R = 0.85), and DeepDTA (R = 0.88). The model uses an ensemble of gradient-boosted and bagged decision trees with 487 interpretable features, requires no GPU, and predicts in under 1 millisecond per complex. We further demonstrate that the framework extends to protein-protein interactions (PPI), achieving MAE = 0.985 kcal/mol on 15 historically "undruggable" PPI targets-outperforming PCANN (1.27), BindPPI (1.40), and PRODIGY (1.4). Applied to microtubule pharmacology, the framework classifies all 24 tested microtubule-targeting agents as stabilizers or destabilizers with 100% accuracy. Finally, we show that MHM principles applied as reasoning middleware to a 7B-parameter large language model improve TruthfulQA accuracy by +4.94 percentage points without model retraining. These results establish Miller Harmonic Mathematics as a unified framework for molecular interaction modeling with demonstrated applications across drug discovery, structural biology, and artificial intelligence.