STRC long-running jobs — bundle 2026-04-22

Two computations to run on Egor’s machine in tmux while Holt-blocked Mini-STRC waits.

What’s in here

FileWhat it doesWall-clock
0-setup.shCreates strc-mmgbsa conda env, installs cas-offinder, downloads hg38 chr15 (PE off-target needs it)15-25 min
1-run-phase4f.shLaunches Phase 4f (interface ΔΔG via OpenMM single-point energies on Ultra-Mini × TMEM145 with top-3 leads pre-bound) in tmuxstarts immediately, runs ~2-6 h
2-run-pe-offtarget.shLaunches Cas-OFFinder off-target scan on the maternal-allele pegRNAs from pe_phase3_allele_discrimination.jsonstarts immediately, runs ~10-30 min
monitor.shtail -f both job logs side-by-sidemanual
pharmacochaperone_phase4f_interface_rescue.pyReal Phase 4f implementation (OpenMM + OpenFF-Toolkit). Replaces the scaffold in ~/Brain/research/strc/models/.
pe_phase4_cas_offinder.pyNew script: reads PE Phase 3 leads, runs cas-offinder against hg38 chr15, gates on off-target count

Run order

cd ~/Desktop/strc-jobs-2026-04-22
 
# One-time setup (long install)
bash 0-setup.sh
 
# Then launch both jobs in detached tmux sessions
bash 1-run-phase4f.sh
bash 2-run-pe-offtarget.sh
 
# Watch progress whenever
bash monitor.sh

Sessions created

tmux sessionWhatLog file
phase4fPharmacochaperone interface-rescue~/Brain/research/strc/logs/phase4f-<timestamp>.log
pe-offtargetPE off-target Cas-OFFinder~/Brain/research/strc/logs/pe-offtarget-<timestamp>.log

tmux ls, tmux attach -t phase4f, Ctrl-b d to detach, tmux kill-session -t phase4f to abort.

Method honesty

  • Phase 4f uses OpenMM single-point energies with implicit GB (GBn2), not full MM-GBSA from MD trajectories. This is a static-snapshot approximation: cheaper, faster, ~1-2 kcal/mol noisier than dynamic MM-GBSA. If the result is unambiguous (clear pass or clear fail), it’s enough. If margin is within ±2 kcal/mol of the 30% gate, we’ll need a follow-up Phase 5 short MD + dynamic MM-GBSA on the top hit.
  • PE off-target uses Cas-OFFinder with up to 4 mismatches on chr15 only (the STRC locus + paralog). For full off-target risk, repeat against full hg38 (~30x larger; not done in this run for time).

Gates

  • Phase 4f passes if ≥1 of {salicylic, nicotinic, cyclopropane-phenyl-COOH} recovers ≥30% of the 8.4 kcal/mol E1659A interface penalty from STRC Electrostatic Analysis E1659A.
  • PE off-target passes if the maternal-allele pegRNA has ≤5 off-target hits with ≤3 mismatches in chr15.

What Egor (parallel session) is doing

While these run, the other Claude session is doing:

  • ASO Phase 1 ViennaRNA fold check on the 18nt candidates
  • Strategy B mRNA-LNP cochlear tropism literature scan (20%/50% OHC gate)