Responder & Non-Responder Analysis

Identify what separates responders from non-responders in Phase 1/2 trial biopsy cohorts, with confounding-adjusted causal inference, multi-layer mechanism recovery, and directionality-aware translational shortlists in a single workflow.

What You Get

Deliverables

Confounding-adjusted responder signature with multi-layer escalation

Multi-method differential expression (limma, DESeq2) with covariate adjustment for trial-design confounders (severity, therapy arm, batch, site). Inverse-probability-of-treatment-weighting and doubly-robust AIPW sensitivity. E-value sensitivity to unmeasured confounding. When single-gene FDR is bound by structural confounding, as it was in PROTECT (severity-therapy collinearity χ²=139.88, zero genes at FDR<0.05 post-adjustment), Inflexa automatically escalates to pathway, TF, network, and immune-deconvolution layers and recovers the signal there.

gene countE-value01,0002,0003,0004,0005,0006,0001.01.11.21.31.41.51.61.71.81.92.02.53.0

Convergent mechanism architecture

Coherent transcriptional programmes recovered from GSEA pathway enrichment, decoupleR TF activity (CollecTRI), WGCNA module hubs, and immune deconvolution, with concordance checked across layers. Programmes get a literature-grounded narrative assembled from PubMed: Inflexa runs targeted PubMed queries against each programme's hub genes, transcription factors, and enriched pathways, traces the evidence chains, and inlines source citations into the dossier. In PROTECT this surfaces as a two-programme architecture: loss of colonocyte identity (Programme A, anchored by SLC26A3, CDX2, HNF1A, FXR) and active inflammatory signalling (Programme B, RELA / STAT3 / NFKB1 / HIF1A drive, CXCL13 lead).

TF activityRELASTAT3HIF1ANFKB1FXR/NR1H4HNF1ACDX1CDX2-8-6-4-20246-7.98-7.56-6.98-6.96+2.85+3.10+3.40+4.23

Directionality-aware tractability triage

For every prioritised candidate, restoration-vs-inhibition direction is computed, then matched against existing chemistry inventories. Counter-mechanistic disqualifications surface before safety review. In PROTECT, FGFR3 inhibitors and pan-CA inhibitors are both counter-mechanistic; FGFR3 is doubly disqualified for pediatric use on growth-plate biology. Multi-mechanism convergence raises priors for restoration targets when independent chemistries hit the same axis: the PPARγ → CDX2 → SLC26A3 axis is hit by rosiglitazone, sodium butyrate, dexamethasone, and FXR agonism.

candidatesRestorationInhibition051015202530T1:1T1:1T2:1T3:2T3:1T4:4T4:1T5:16T5:3T1T2T3T4T5
DECISION ENABLED

Move from a confounded responder cohort to a defensible biomarker, MoA, and tractability shortlist your translational team can act on: restoration targets pursued with the right chemistry, counter-mechanistic disqualifications removed up front, mechanism evidence chains traced to PubMed source for internal review.

Sample Output

PROTECT pediatric UC re-analysis: confounding, programmes, and tractability

PROTECT (GSE109142): Pediatric UC, week-4 remissionn=206
206
PATIENTS (105 R / 101 NR)
21,004
GENES SCREENED
139.88
Χ² (THERAPY × SEVERITY)
688
CONCORDANT PATHWAYS
30
PRIORITISED CANDIDATES
24/30
REQUIRE RESTORATION
Therapy × Severity confounding structureχ²=139.88, p<0.0001
TherapySeverity5-ASACS-POCS-IVLowModerateHigh5328391461947

5-ASA was prescribed exclusively to Low-severity patients. CS-IV concentrated in High-severity disease. Single-gene signal is structurally bounded; the agent escalates.

Genome-wide E-value distributionmedian 1.19, 4 genes ≥ 3
gene countE-value01,0002,0003,0004,0005,0006,0001.01.11.21.31.41.51.61.71.81.92.02.53.0

Only 4 genes reach E ≥ 3, the threshold for ROBUST_STRONG. When the single-gene signal is confounding-bound, the platform escalates to pathway, TF, and network layers.

Pathway concordance: naïve DE vs IPTW NES688 concordant, 0 discordant
IPTW NESnaïve DE NES-3.5-3-2-101233.5-3.5-3-2-101233.5HALLMARK_OXPHOSHALLMARK_TNFA_NFKBIL6_JAK_STAT3REACTOME_COMPLEMENTHALLMARK_EMT

Pathway-level signal survives confounding adjustment. Coherent programmes emerge from convergent layers.

Transcription factor activity (CollecTRI ULM)responder vs non-responder
TF activityRELASTAT3HIF1ANFKB1FXR/NR1H4HNF1ACDX1CDX2-8-6-4-20246-7.98-7.56-6.98-6.96+2.85+3.10+3.40+4.23

Two-programme architecture: colonocyte-identity restoration drivers (CDX2, CDX1, HNF1A, FXR) above zero; inflammatory-signalling drivers (RELA, STAT3, NFKB1, HIF1A) below.

Directionality-aware tractability triage30 candidates: 24 restore / 6 inhibit
candidatesRestorationInhibition051015202530T1:1T1:1T2:1T3:2T3:1T4:4T4:1T5:16T5:3T1T2T3T4T5
CandidateDirectionTierNote
FGFR3RestorationT1Counter-mechanistic: inhibitors disqualified
SLC26A3RestorationT2Inhibitor-only chemistry; needs activator series
CDX2RestorationT5Restored via PPARγ → CDX2 axis (rosiglitazone, butyrate)
CXCL13InhibitionT4Biologic only; no clinical-stage agent in IBD
RELAInhibitionT5Programme B driver; hit by JAK1 inhibition (upadacitinib)

PROTECT is one example. The full case study traces the confounding adjustment, the two-programme architecture, the directionality framework that disqualifies FGFR3 inhibitors entirely, and the pediatric translational re-ranking that surfaces sodium butyrate, upadacitinib, and anti-CXCL13 as the three actionable programmes. Read the full PROTECT case study → /news/protect-pediatric-uc-responder-analysis

How It Works

Methodology

STEP 1

Cohort design assessment + confounding triage

Inflexa inventories design covariates (severity, therapy arm, batch, site, demographics), runs the contingency tests, and flags structural confounding before analysis. In PROTECT this immediately surfaced χ²=139.88 between therapy and severity, bounding what single-gene analysis could discover.

STEP 2

Multi-method DE with sensitivity

limma and DESeq2 with covariate adjustment for the flagged confounders. Inverse-probability-of-treatment-weighting and doubly-robust AIPW sensitivity. E-value computation per gene against unmeasured confounders, returning a genome-wide robustness profile.

STEP 3

Multi-layer enrichment escalation

GSEA against Hallmark, Reactome, and custom collections; decoupleR TF activity (CollecTRI ULM); WGCNA modules with eigengene-trait correlation; immune deconvolution. Cross-layer concordance check identifies the coherent programmes that survive when single-gene signal does not.

STEP 4

PubMed-curated MoA narrative

For each programme, Inflexa runs targeted PubMed queries against the programme's hub genes, transcription factors, and enriched pathways. Literature evidence chains are assembled (basic → preclinical → clinical) and inlined into the dossier with full provenance to PMID.

STEP 5

Tractability triage with directionality + dossier assembly

Direction-of-effect per candidate; tractability tiering against ChEMBL and DrugBank inventories; multi-mechanism convergence layered on; directionality-mismatched candidates flagged. Full dossier assembled with provenance, ready for translational, clinical, and external review.

Who This Is For

Target personas

Translational medicine lead

Build a defensible responder/non-responder story from a Phase 1/2 cohort, with convergent mechanism evidence and a tractability-triaged shortlist your discovery and safety teams can act on.

Exploratory biomarker scientist

Recover a coherent biomarker signature when single-gene FDR is bound by trial-design confounding. Multi-layer escalation and convergent evidence stacks beat bigger gene lists.

Clinical pharmacologist

Couple responder transcriptomics to mechanism and direction-of-effect, so candidate triage reflects what restoration-vs-inhibition pharmacology can actually achieve in patients.

Related Use Cases

Explore more

Bring us a Phase 1/2 cohort.

Tell us about a translational dataset you care about. We'll scope a paid discovery collaboration in 2–3 weeks and run a full analysis with weekly check-ins.