Case Study · Melanoma IO Resistance · Deep Dive

Anti-PD-1 Resistance in Melanoma: What 133 GaiaLab Analyses Reveal

Why does immunotherapy resistance happen? GaiaLab aggregated 133 analyses of the melanoma IO resistance gene panel to identify recurring mechanisms, validated drug directions, and open questions that remain unresolved.

133

Analyses of melanoma IO resistance panels

Gene panel: checkpoint + resistance genes analyzed

Recurring resistance mechanisms identified

0.544

Platform AUROC (95% CI: 0.526–0.562)

Gene Panel Analyzed

The Melanoma IO Resistance Panel

This analysis uses a 10-gene panel covering both checkpoint pathways (immunotherapy targets) and resistance mechanisms (why those targets fail). The panel is pre-configured in the GaiaLab gallery.

PDCD1 CD274 CTLA4 LAG3 HAVCR2 PTEN B2M JAK1 BRAF STK11

Teal = checkpoint / IO pathway. Red = resistance mechanism gene.

Resistance Biology

Key Genes Driving IO Resistance

Across 133 analyses, four resistance genes consistently emerged as primary drivers. Here is the mechanism each one contributes:

PTEN loss

PTEN deletion activates the PI3K/AKT pathway, which suppresses T-cell infiltration and reduces tumor immunogenicity. Associated with primary resistance to anti-PD-1 therapy in multiple studies.

B2M mutation

B2M (Beta-2 microglobulin) loss disrupts MHC class I surface expression, preventing CD8+ T cells from recognizing and killing tumor cells regardless of PD-1 blockade status.

JAK1/JAK2 loss

JAK1 or JAK2 inactivating mutations disrupt IFN-gamma signaling — the pathway required for PD-L1 upregulation and T-cell cytotoxicity. Cells become functionally invisible to checkpoint blockade.

STK11 mutation

STK11 (LKB1) loss creates a non-T-cell-inflamed "cold" tumor microenvironment. Associated with resistance to both anti-PD-1 monotherapy and combinations in BRAF-mutant melanoma subsets.

Live Resistance Prediction

Run resistance prediction for Pembrolizumab with this resistance panel (PTEN, B2M, JAK1, STK11) — see ranked mechanisms, confidence scores, and preemptive combination strategies.

Run resistance prediction →

Validated Drug Directions

Checkpoint Combinations With Confirmed Trials

Both Pembrolizumab and Ipilimumab were scored as validated candidates in the 133-analysis aggregate. Each has multiple confirmed clinical trials in melanoma:

Anti-PD-1 monoclonal antibody targeting PDCD1/PD-1. Primary checkpoint blockade in melanoma. GaiaLab identified it as validated across all 133 analyses given the PDCD1 and CD274 genes in panel. Combination strategies with LAG3 inhibitors are an active research direction captured in the emerging targets section below.

NCT04068181 NCT01295827 NCT03715205

Anti-CTLA-4 monoclonal antibody targeting CTLA4. Second major checkpoint pathway in melanoma. In combination with anti-PD-1 agents, it provides orthogonal checkpoint blockade — critical for tumors where a single checkpoint axis is insufficient due to PTEN or JAK1 loss.

NCT00094653 NCT03724968 NCT00790010

All 6 NCT IDs verified against ClinicalTrials.gov API v2 · 2026-05-22. No hallucinated trials found.

Emerging Targets

Beyond PD-1 and CTLA-4

GaiaLab's analysis of LAG3 and HAVCR2 (TIM-3) flagged both as emerging resistance targets with active trial activity. These represent the next generation of checkpoint strategies for IO-resistant disease:

LAG3

Relatlimab (LAG3 inhibitor)

LAG3 is a co-inhibitory receptor expressed on exhausted T cells. Relatlimab + nivolumab (Opdualag) received FDA approval in 2022 for unresectable melanoma — a validation of the LAG3 target hypothesis. GaiaLab identifies this axis in 89% of melanoma IO resistance panels.^†

† Computed from GaiaLab analyses containing the melanoma IO resistance panel (LAG3 + PDCD1 + CTLA4 genes), May 2026. Local snapshot data shows relatlimab/LAG3 surfaced in 100% of 55 verified melanoma analyses; 89% reflects the broader production dataset including Railway DB analyses.

HAVCR2 (TIM-3)

TIM-3 inhibitors (multiple in trial)

TIM-3 is expressed on dysfunctional T cells and is co-expressed with PD-1 in melanoma. Multiple TIM-3 inhibitors are in Phase I/II trials. GaiaLab consistently flags HAVCR2 in IO resistance panels as an unvalidated but mechanistically plausible target.

Model Performance

Platform Accuracy

GaiaLab's global AUROC is 0.544 (bootstrap 95% CI: 0.526–0.562) vs 0.50 random baseline, computed across all resolved predictions in the ledger. Disease-specific scores are not yet computed with sufficient sample size per disease to report individually — the figures will be published on the validation page as the ledger grows. Prospective tracking began May 2026; retrospective co-occurrences are used for current calibration.

Metric	Value	vs baseline
Global AUROC (all diseases)	0.544
Brier score (lower = better)	0.186
No-skill AUROC baseline	0.500

Full calibration data and Brier curve: validation page · calibration chart.

Honest Assessment

What We Don't Know Yet

GaiaLab is a hypothesis-generation system. The following questions are outside what the current data can resolve:

No completed outcome data exists for resistance-specific checkpoint combinations (e.g., anti-PD-1 + LAG3 + PTEN-restoration strategies) — all active or enrolling.
The causal direction of STK11 and PTEN loss relative to IO resistance is associative in current literature; GaiaLab does not claim these are experimentally confirmed as causal in melanoma-specific models.
Whether TIM-3 inhibitors will show benefit in HAVCR2-high melanoma subsets is unresolved — Phase II readouts are pending as of the analysis date.
Combination sequencing (which checkpoint to target first, which second) is not something GaiaLab currently models — it scores drugs in isolation or pairwise, not as ordered regimens.
Patient-level biomarker thresholds (what level of B2M expression predicts resistance vs. sensitivity) are not computed by GaiaLab — that requires matched tumor data outside the platform's current scope.

Analyze your melanoma panel

Run the full 10-gene melanoma IO resistance analysis with live data from 75+ biological databases. Results in under 60 seconds.

Run melanoma IO resistance analysis →