What question did this study set out to answer?

The aim is to validate the Luoshu-Markov framework through three key biochemical tasks.

May 15, 2026Open Access

Luoshu-Markov Framework: Three-Task Validation with Honest Positioning — Allosteric Benchmark (n=293), Multi-Target Vina Discrepancy (8 targets), and Calibrated State-of-the-Art Comparison

Key Points

The aim is to validate the Luoshu-Markov framework through three key biochemical tasks.
Conducted allosteric benchmarking on ASBench+ASD-90 with n=293 samples.
Evaluated multi-target Vina discrepancies across 8 protein families with n=46 ligands.
Performed calibrated comparisons of the framework against existing benchmarks.
Achieved 5-fold CV AUC=0.68 for allosteric benchmark, compared to PASSer's 0.72.
+0.03 AUC improvement noted with lambda_2 over CHC alone.
The framework did not report superiority over PASSer/ALLO/AlloBench in calibrated comparisons.

Abstract

An honest, calibrated validation of the Luoshu-Markov framework across three tasks. (1) Allosteric benchmark on ASBench+ASD-90 (n=293, 207 positive / 86 negative): 5-fold CV AUC=0. 68 comparable to PASSer (0. 72). lambda₂ adds +0. 03 AUC over CHC alone. (2) Multi-target Vina discrepancy in 8 protein families (n=46 ligands): 37/46 under-scored by Vina. The lambda₂ domain-boundary explanation holds for kinases (CDK2, EGFR, BRAF, ABL1) but fails for HIV-PR (symmetric dimer) — an honest negative finding. (3) Calibrated comparison: this framework is NOT claiming to be the world-first in any component; the novelty is the C-alpha-only Lo-Shu+Markov-Fiedler integration covering three tasks in one lightweight pipeline. Current status: promising proof-of-concept warranting formal benchmark comparison vs PASSer/ALLO/AlloBench on a common test set.

AI에게 질문

Bookmark

View Full Paper