Charge-sign asymmetry in protein crystallization outcomes across 52,307 purified structural-genomics targets (TargetTrack, 2017 release)

Abstract

Statistical analysis of protein crystallization outcomes in the final TargetTrack release (1 July 2017; 52,307 purified structural-genomics targets across 17 centres, within-centre models). Finding: at equal net-charge magnitude, positively charged proteins are recorded as crystallized less often than negatively charged ones. With the optimum at about -22 charges per 1,000 residues, the odds ratio at q = +40 is 0.441 [0.407, 0.48] versus 0.915 [0.891, 0.94] at q = -40, an asymmetry ratio of 2.08 (nested likelihood-ratio test for sign asymmetry, with the symmetric model free to choose its own centre on a grid: LR = 38.3, p = 9.9e-7; the less conservative plug-in-centre version gives LR = 87.3). The direction is unchanged after adjusting for side-chain conformational entropy (ratio 2.08 to 1.88), after excluding nucleic-acid-binding and membrane-associated families (2.08 to 2.15), and when restricted to the centre sets of stricter failure definitions (2.04-2.06); across four definitions of crystallization failure the ratio spans 1.90-3.39 and the direction is robust, but the magnitudes are not comparable across denominators. In the subset with per-residue disorder predictions (n = 8,397-8,481) the effect is carried by the folded region rather than the disordered one. This is a historical-cohort association, not a demonstration of an electrostatic mechanism; a sign asymmetry in surface charge was previously reported by Fusco et al. 2014 (PLoS ONE 9:e101123) on a different estimand, so this work is a population-scale quantification rather than a first report. Not yet done: disorder adjustment on the full cohort, which is the one outstanding item that can change the headline number. The package contains the analysis manifest, derived tables, and build and verification scripts. Third-party screening-formulation data are not redistributed: component-level and per-well tables are excluded and replaced by per-generation summary statistics. Excluded files are listed with their SHA-256 and provenance in DEPOSIT_EXCLUDED.txt. Files restricted pending publication.