Reliability — honest assessment of every hypothesis

Each hypothesis is scored 0–100% based on:

• Number of independent validation panels
• Convergence at increasing docking exhaustiveness
• Specificity of negative controls (random + within-class + chemical analogue)
• Mechanistic plausibility
• Known wet-lab evidence (when present)

For H1 (phenytoin selectivity), our reliability score reflects:

• The selectivity signal is well above the noise floor (Δ = −0.58 / noise = 0.5 → ratio ~1.2×)
• The signal is convergent across exhaustiveness (−0.58 to −0.61)
• The discrimination is real (0/7 random control)
• 17/19 AEDs do not show the signal (specific within-class)
• The pharmacophore explains why
• Phenytoin is used empirically in PACS1 with anecdotal good response

What it does NOT mean:

• It does NOT mean phenytoin is guaranteed effective in clinical trials
• It does NOT mean other AEDs cannot help — they may help via canonical mechanisms
• It does NOT mean the docking captures the true binding pose (this requires X-ray or cryo-EM)

• All results are in silico. ~±0.5 kcal/mol uncertainty in Vina even at exhaust=32.
• ESMFold receptor structure is not experimentally validated (no PDB structure of PACS1-FBR).
• AlphaFold 3 was used with single seed and isolated complexes (no membrane, no cofactors).
• Pharmacophore based on 25 negative + 3 positive analogues — broader testing would refine.
• MD production simulation (100 ns × 3 replicas) set up but not executed (requires GPU).
• No wet-lab experimental validation has been done — this is the critical next step.

In the broader scientific literature, reliability is often opaque ("This drug is a candidate for X disease."). Our open scoring lets a reader decide how much weight to put on each statement. The pocket #19 downgrade (from initially 65% to final 30%) is a documented example of this transparency.