🏠 Home 🎯 Heroes 🧬 Gene Scan 🧬 Single-Cell πŸ“Š Graphs 🧠 NPEAD βš™οΈ Ops

🧬 Single-Cell Intelligence

Cell-type resolved target characterization β€” understanding WHERE each drug target lives in the tumor microenvironment and WHY it matters for therapy.

966
Genes in Analysis Pipeline
4
Cancer Types
6
Cell Compartments
3
Resolution Layers

Why Single-Cell Matters for Drug Discovery

The problem bulk data can't solve

CortexQ uses single-cell data to determine: (1) Is the target in tumor cells, stroma, or immune? (2) Is it homogeneous or heterogeneous? (3) Does expression change after treatment?

Single-Cell Processing Architecture

scRNA-seq datasets (GEO/CellxGene/TCGA) β†’ Quality filtering (cells: >500 genes, <20% mito) β†’ Normalization (scran/Seurat) β†’ Cell type annotation (reference-based: Azimuth/scType) β†’ Per-gene analysis: β”œβ”€β”€ Tumor specificity (Ο„ score: expression in tumor vs all other cells) β”œβ”€β”€ Compartment score (tumor / stroma / immune / endothelial / neural / mixed) β”œβ”€β”€ Heterogeneity index (% of tumor cells expressing target) β”œβ”€β”€ Treatment response (expression change preβ†’post therapy) └── Resistance population (is target upregulated in resistant clone?) β†’ Feed into NPEAD V2 (disease context dimension) β†’ Feed into CortexQ decision (modality selection, combination logic)

Cell Compartments & Target Implications

CompartmentCell TypesIf Target Is HEREImplication for Drug Design
πŸ”΄ TumorMalignant epithelial, cancer stem cellsDirect kill targetSmall molecule / ADC / PROTAC β€” kill the cancer
🟑 StromaCAFs, fibroblasts, stellate cellsMicroenvironment targetTGF-Ξ² trap, FAP-targeted β€” remove barrier
πŸ”΅ ImmuneT cells, macrophages, DCs, MDSCsImmune modulation targetCheckpoint, CSF1R, BiTE β€” reprogram immunity
🟒 EndothelialBlood vessels, lymphaticsVascular targetAnti-VEGF, vascular normalization
🟣 NeuralNeurons, Schwann cellsNerve-cancer interactionRare β€” perineural invasion targets
βšͺ MixedExpressed in multiple compartmentsSelectivity problemNeed tumor-selective delivery (ADC) or conditional activation

Per-Gene Single-Cell Metrics

MetricRangeWhat It Tells CortexQ
Tumor Specificity (Ο„)0-1Ο„ > 0.8 = highly tumor-specific. Ο„ < 0.3 = expressed everywhere (toxicity risk)
Compartment ScoreCategoryWHERE the target lives β†’ determines modality and combination logic
Heterogeneity Index0-100%>90% of tumor cells express β†’ homogeneous (good for monotherapy). <50% β†’ heterogeneous (ADC or combo needed)
Immune Cell Fraction0-1If target is >60% in immune cells β†’ immune modulation target, not direct kill
Treatment Delta-/0/+Target goes UP after treatment β†’ resistance marker. Goes DOWN β†’ already being hit.
Resistance Clone FlagY/NIs the target specifically enriched in drug-resistant subpopulation?

Cancer-Specific Single-Cell Context

CancerKey Question sc AnswersCritical Cell TypesDatasets
πŸ”΄ MSS CRCWhy are T cells excluded? Which cells produce TGF-Ξ²? Where are MDSCs?M2 macrophages, CAFs, Tregs, CMS4 mesenchymalLee et al 2020, Pelka et al 2021
πŸ”΅ ProstateWhich cells drive castration resistance? Where does lineage plasticity start?Luminal progenitors, neuroendocrine, AR-low subpopChen et al 2021, He et al 2021
🟒 LungWhich cells have driver mutations? What happens after TKI resistance?AT2-like cancer cells, EMT intermediate, SCLC-transformedKim et al 2020, Maynard et al 2020
🟣 PancreaticHow dense is the stroma? Which CAF subtype dominates? Are T cells even present?myCAFs, iCAFs, stellate cells, ductal malignantPeng et al 2019, Steele et al 2020

Integration with CortexQ Pipeline

How single-cell data feeds into decisions
Decision Pointsc Data UsedExample
Modality SelectionCompartment scoreTarget in immune cells β†’ checkpoint/CSF1R, NOT small molecule kill
Combination DesignResistance clone flagTarget upregulated in resistant clone β†’ add as combo partner
Therapeutic WindowTumor specificity (Ο„)Ο„ < 0.5 β†’ target in normal tissue too β†’ toxicity risk β†’ need selective delivery
Patient SelectionHeterogeneity index<50% of cells express target β†’ only works in subset β†’ need biomarker
NPEAD V2 ScoringDisease context dimensionCompartment + heterogeneity feed into disease-context score
Disproof ExperimentTreatment deltaIf target goes DOWN after chemo β†’ it's not a resistance target β†’ deprioritize

Processing Pipeline (sc_processor.py)

StepMethodOutput
1. Data AcquisitionGEO/CellxGene API downloadCount matrices + metadata
2. QC Filtering>500 genes/cell, <20% mito, doublet removalClean cell Γ— gene matrix
3. Normalizationscran pooling + log-transformNormalized expression
4. Cell AnnotationReference mapping (Azimuth/scType)Cell type labels per cell
5. Target ScoringPer-gene: Ο„, compartment, heterogeneityGene-level metrics β†’ DDB
6. Resistance AnalysisPre/post treatment differentialTreatment delta + clone flags
7. NPEAD IntegrationMetrics β†’ disease context dimensionFeeds nightly scoring

Key IO Heroes β€” Single-Cell Context

GeneCompartmentPrimary Cell TypeImplication
CSF1RπŸ”΅ ImmuneM2 macrophages (TAMs)Block to reprogram myeloid β†’ sensitize to checkpoint
TBK1πŸ”΄ Tumor + πŸ”΅ ImmuneTumor cells (silenced STING) + DCsReactivate innate sensing in tumor β†’ recruit T cells
IDO1πŸ”΅ Immune + 🟑 StromaDCs, MDSCs, CAFsRemove metabolic suppression β†’ T cells regain function
TGFBR1🟑 StromaCAFs (myofibroblastic)Block stromal barrier β†’ allow T cell infiltration
FAP🟑 StromaCancer-associated fibroblastsTarget CAFs specifically (tumor cells FAP-negative)
CD274 (PD-L1)πŸ”΄ Tumor + πŸ”΅ ImmuneTumor surface + macrophagesBoth compartments express β†’ checkpoint blockade hits both

Current Status

ComponentStatusDetails
sc_processor.pyβœ… BuiltFull pipeline: QC β†’ annotation β†’ scoring β†’ DDB write
CRC datasets (Pelka et al)🟑 Queued366K cells, 32 patients, pre/post immunotherapy
Prostate datasets (Chen et al)🟑 QueuedmCRPC, 14 patients, AR signaling resolved
NPEAD V2 integrationβœ… LiveDisease context dimension accepts sc metrics
Compute🟑 On-demandc5.4xlarge ($0.68/hr), ~30 min/cancer type

Single-cell data tells CortexQ WHERE targets live. Bulk data only tells IF they're expressed.
The difference is the difference between a drug that works and one that causes toxicity.
MindNetQ Inc. Β· Updated July 11, 2026