validate.py

"""Validate our port reproduces the original model's saved prediction exactly.

PichiaArch1_HerHC.xlsx was generated by the reference apply-notebook using the
SAME checkpoint we deployed (2Target_AllData/FinArch1_AttnCorr_cp). We recover
the input protein from that ground-truth output, run it through our model, and
compare codon-for-codon.
"""
import pandas as pd
import model as clm

BASE = (r"C:\Users\ChemEGrad2025\MIT Dropbox\Love Lab\Love Lab Member Archives"
        r"\Harini Narayanan April 2022 - July 2025\CodonOptimization\CO_Application")
REFS = {
    "Herceptin HC": BASE + r"\Herceptin_HC\PredictedSequences\PichiaArch1_HerHC.xlsx",
    "Herceptin LC": BASE + r"\Herceptin_LC\PredictedSequences\PichiaArch1_HerLC.xlsx",
    "Lactoferrin":  BASE + r"\Lactoferrin\PredictedSequences\PichiaArch1_Lactoferrin.xlsx",
}

CODON_TO_AA = {c: aa for aa, cs in clm.DIC_AA_CODON.items() for c in cs}
opt = clm.CodonOptimizer()

for name, path in REFS.items():
    row = pd.read_excel(path).iloc[0].tolist()
    ref_tokens = [int(t) for t in row[1:] if pd.notna(t)]   # drop index column
    assert ref_tokens[0] == clm.CODON_START and ref_tokens[-1] == clm.CODON_STOP
    ref_dna = clm.detokenize_codons(ref_tokens[1:-1])       # strip 65 / 66
    ref_codons = [ref_dna[i:i+3] for i in range(0, len(ref_dna), 3)]

    aa_input = "".join(CODON_TO_AA[c] for c in ref_codons)  # recover protein (+'*')
    our_dna = opt.optimize(aa_input)
    our_codons = [our_dna[i:i+3] for i in range(0, len(our_dna), 3)]

    exact = our_dna == ref_dna
    n = min(len(ref_codons), len(our_codons))
    same = sum(a == b for a, b in zip(ref_codons[:n], our_codons[:n]))
    print(f"{name:14s} {len(aa_input.rstrip('*')):4d} aa | ref {len(ref_codons)} / "
          f"ours {len(our_codons)} codons | agree {same}/{n} "
          f"({100*same/n:.2f}%) | EXACT={exact}")