Design Experiments From Reported Results

 Long Analysis Plan Copied

Testable axis: interface perturbation → functional deficit → severity rank

The strongest scaffold for this plan is the EXOSC3/RRP40 study, which systematically modeled eight patient missense variants in yeast Rrp40 and mouse EXOSC3, combining AlphaFold2/3-multimer structures with Rosetta Cartesian ΔΔG calculations, and found that steady-state protein loss alone did not predict phenotype — variant-specific interface and cofactor-interaction defects did EXOSC3/RRP40 variant-specific functional consequences study. Reported quantitative anchors to build the genotype–phenotype axis from: Rrp40-Y64N at 20% of WT steady-state protein, G148C at 29%, W195R at 53%, yet G148C showed no growth or genetic-interaction defect despite a protein level similar to W195R, which was temperature-sensitive Variant protein-level vs phenotype dissociation. This dissociation is exactly why an interface-perturbation score, not abundance, should be the predictor variable.

Step 1 — Computational prediction layer (per allele)

For each patient variant: (a) build homo-dimer and complex models with AlphaFold2/AlphaFold3-multimer exactly as in the source study Computational toolchain of the source study; (b) compute per-variant ΔΔG with Rosetta Cartesian ΔΔG plus orthogonal tools — the IMPG2 case shows cross-tool consistency is measurable, with Arg291Ser yielding ΔΔG of −1.02 (DynaMut2), −1.276 (mCSM), and −1.11 (MuPRO), all destabilizing while sequence-based tools (SIFT 0.09 tolerated, PolyPhen2 0.242 benign) were uninformative IMPG2 stability-prediction tool discordance; (c) compute an interface-perturbation index = ΔΔG at oligomer-interfacial residues + predicted change in interface buried surface area + co-IP-observable contact loss (EXOSC8/9/10 binding was markedly reduced for Y108N in mouse N2a cells Interface contact-loss evidence (co-IP)). Predict severity rank a priori, register predictions before assays.

Step 2 — Patient-variant protein engineering

Generate each variant by site-directed mutagenesis (as done for EXOSC3 via QuickChange II and CRISPR genome editing, and for IMPG2 designs via QuikChange in the EPO study workflow Engineering-and-assay workflow precedent). Add a first-pass purification-yield stability screen: in the CobT mutational series, yield tracked functional loss — WT 5 mg/g paste, G257D 0.03 mg/g (a ~99% kcat/Km loss), G171D 0.2, G320D 0.1 mg/g — making yield a cheap, quantitative early readout of misfolding/stability CobT yield-vs-function stability proxy.

Step 3 — Allele-specific stability and dimerization assays

For every engineered allele, run: (i) DSF/nanoDSF and CD melting curves for per-allele thermal stability; (ii) SEC-MALS or AUC to quantify oligomer/dimerization propensity of hinge/head-like domains — mirroring the EXOSC3 complex-assembly question Complex-assembly and genetic-interaction readouts; (iii) quantitative co-IP / pull-down binding curves to the partner subunits; (iv) RNA-processing functional readout (7S pre-rRNA accumulation in yeast as the functional anchor Functional RNA-processing readout).

Step 4 — Quantitative genotype–phenotype test

Regress the measured interface-perturbation index (ΔΔG at interface + binding-affinity loss + assembly defect) against cognitive-severity rank across carriers using Spearman rank correlation with permutation-based p-values (the oncology pipeline precedent uses similar permutation/hypergeometric statistics and achieved up to 93.17% PPV at top-percentile ranking, demonstrating that computed perturbation scores can quantitatively stratify clinical severity Perturbation-score-to-clinical-outcome statistics precedent). Falsification: if interface-index rank does not outpredict both protein abundance (which failed: 29% G148C with no defect vs 53% W195R with defect) and sequence-based scores, the axis collapses.

Blindspots and uncertainties

• Yeast/animal-cell models lack patient-derived cells or vertebrate in-vivo validation; no transcriptome-wide profiling was performed Model-system validation gap.
• Severity ranking across carriers is confounded by variant zygosity, compound heterozygosity, and ascertainment bias (single-patient IMPG2 case had no segregation data Segregation and sample-size limitations).
• In-silico binding predictions have documented discordance with ex vivo outcomes (EPO: only EPO-4.1 showed notable predicted binding decrease while five mutants reduced T-cell responses) In-silico vs wet-lab discordance evidence — hence experimental measurement, not prediction, must define the final index.