Machine learning pipelines for Multiple Sclerosis transcriptomics analysis can leverage post-quantum cryptographic techniques to securely process and share sensitive genetic data across network stacks.

Two independent, falsifiable sub-claims are bundled in this discovery and must be tested separately:

H1 (Cryptography/Systems claim): A post-quantum cryptographic (PQC) protocol stack (e.g., CRYSTALS-Kyber for key encapsulation + CRYSTALS-Dilithium for signatures, or equivalent NIST-standardized primitives) can be integrated into a multi-institutional MS transcriptomics ML pipeline (scVI training, DEG computation, network-proximity scoring) such that (a) end-to-end encrypted data transfer/storage adds ≤20% wall-clock overhead and ≤2× storage overhead versus classical TLS 1.3/AES-256 baselines, and (b) no ML pipeline output (DEGs, composite scores, cluster assignments) changes by more than stochastic re-run variance (defined as <5% relative difference in log2FC and <0.01 absolute difference in composite score) when data passes through the PQC-secured channel versus an unencrypted control.

H2 (Biological claim, inherited from internal analysis, unreplicated): DNMT1, ZNF740/BRD3, and CTSS are causally implicated in CD8+ T-cell-driven CA-RIM (chronic active rim) pathology in smoldering MS, such that pharmacological inhibition (azacitidine/decitabine for DNMT1; JQ1/birabresib for ZNF740-BRD axis; RO5459072 for CTSS) measurably reverses the CA-RIM effector transcriptional signature in CD8+ T cells in vitro and/or in vivo.

H1 and H2 are logically independent: H1 could be fully true with H2 entirely false, and vice versa. This EVP validates them on separate tracks.

Track A (H1 — Crypto, minimum viable test, 4–6 weeks):

1. Stand up 2-node simulated multi-institutional environment (site A: GEO data host; site B: compute/analysis host).
2. Benchmark classical TLS 1.3 baseline: transfer GSE193770 + GSE138614 (~15–40 GB raw fastq/h5ad) and run full Phase 1–4 pipeline; record wall-clock, CPU/GPU hours, checksums of DEG/composite-score outputs.
3. Replace transport layer with PQC hybrid handshake (Kyber768 KEM + Dilithium3 signatures via liboqs-OpenSSL provider); repeat identical pipeline run.
4. Diff outputs bit-for-bit where deterministic (DEG tables, STRING scores) and statistically (scVI stochastic runs, n=5 seeds each condition).
5. Record overhead deltas; run at 3 payload sizes (1GB, 10GB, 40GB) to characterize scaling.

Track B (H2 — Biology, minimum viable test, 10–14 weeks):

1. Obtain/confirm access to CD8+ T cells from ≥15 CA-RIM MS patients + 15 controls (banked PBMC/CSF or new IRB-approved collection).
2. FACS-sort CD8+ T cells; validate DNMT1, ZNF740, CTSS expression via RT-qPCR and targeted scRNA-seq against pipeline predictions.
3. Culture sorted CD8+ T cells under CA-RIM-mimicking stimulation (IFN-γ/TCR activation); treat with azacitidine (0.1–1 µM), JQ1 (0.1–1 µM), RO5459072 (0.01–1 µM) vs. vehicle, 48–72h.
4. Measure effector program readout (IFN-γ ELISA/intracellular flow, RNA-seq of treated vs. control) and CTSS enzymatic activity (fluorogenic substrate assay) and MBP-cleavage functional assay.
5. Compare against composite score predictions and network proximity rankings.

# TRACK A: PQC-secured pipeline benchmark
for condition in [classical_tls, pqc_hybrid_kyber768_dilithium3]:
    for payload_gb in [1, 10, 40]:
        establish_channel(condition)
        t0 = now()
        transfer(GSE193770, GSE138614, channel=condition)
        run_phase1_deg()          # 1,065 DEG check
        run_phase2_scvi(seeds=5)  # 32,239-cell atlas, NOT 36,966
        run_phase3_composite_score()
        run_phase4_string_proximity()
        t1 = now()
        log(condition, payload_gb, latency=t1-t0, output_hash=checksum(results))
compare_outputs(classical_tls_outputs, pqc_hybrid_outputs)
assert relative_diff(log2FC) < 0.05
assert abs_diff(composite_score) < 0.01
assert (t1-t0)_pqc <= 1.20 * (t1-t0)_classical

# TRACK B: wet-lab target validation
cohort = recruit_CA_RIM_patients(n=15) + controls(n=15)
cd8_cells = FACS_sort(cohort, marker="CD8+")
baseline_expr = RTqPCR(cd8_cells, genes=["DNMT1","ZNF740","CTSS"])
for drug, target in [(azacitidine, DNMT1), (JQ1, ZNF740_BRD3), (RO5459072, CTSS)]:
    treated = culture_and_treat(cd8_cells, drug, doses=[0.01,0.1,1.0], hrs=72)
    readout = measure_IFNg_effector_program(treated) + functional_assay(target)
    dose_response_fit(readout)
compare_to_pipeline_predictions(composite_scores, network_proximity)