Stage 05

Pose persistence in molecular dynamics

The fifth stage tests whether a predicted ligand-receptor pose survives in explicit solvent.

Residue fluctuations of EcDHFR with a bound ligand, highlighting the M20 loop region
Residue fluctuations (RMSF) from the trajectory. Stabilisation of the M20 loop is what differentiates analogues.

System preparation#

The EcDHFR receptor (PDB 6XG5) is stripped of alternative conformations, water and buffer. The protein is described with the AMBER ff14SB force field and TIP3P water; the ligand is parameterised with GAFF2 / AM1-BCC through ACPYPE. The starting pose comes from the predicted Boltz-2 complex.

Simulation and read-out#

Production dynamics runs for 0.1 ns per compound. From the trajectory we derive a backbone deviation and a stability indicator, and report a stable/unstable verdict. We do not use it as a gate: only 3.85% of runs come out stable, and at this length with a single replica the verdict would carry no statistical power. Results from this stage contain only quantities with an unambiguous physical interpretation.

Limitations#

A single short trajectory settles neither energetic stability nor the ligand's effect on hydride transfer. That would require longer replicas, free-energy methods or QM/MM calculations. Because of CUDA library incompatibilities on the shared cluster, we shorten the dynamics by default, which lowers confidence in its metrics further.

The measured cost is about 46 s per molecule (median 41 s) with a large standard deviation that reflects the variability of convergence, or roughly 33% of a reference 50-molecule run.