PhD Thesis — Mohammad Ali | Monocyte→Macrophage Master Regulators

Project brief

What this PhD aims to achieve

This computational thesis studies how human monocytes become macrophages. By integrating gene-expression and epigenomic data, it will reconstruct the gene-regulatory network governing this cell-fate transition and identify the small set of transcription factors that act as master regulators.

Scientific objective

Build a validated, multi-omics model of monocyte→macrophage differentiation.
Rank and test the transcription factors most likely to causally control macrophage fate.

End deliverables

A reproducible regulatory-network model and validated master-regulator list.
In-silico knockout predictions, immunotherapy target hypotheses, and one proposed wet-lab validation experiment.

Clinical relevance: the monocyte→macrophage axis is used as a model to nominate immunotherapy targets for cancers and tumor microenvironments where CAR-T therapy has limited effectiveness.

Step 0 — Before the thesis begins

🎓 Proposal Defense — Thesis Gate

Before any analysis begins, the research proposal must be written, submitted to the committee, and defended. This gate must be cleared before Phases 1–10 are executed. The pipeline design and dataset strategy described below will form the core of the proposal.

Thesis phases

Data Availability Scouting

Survey GEO, Human Cell Atlas, ENCODE, BLUEPRINT

Completed ▶

What was done

Searched GEO, Human Cell Atlas (CellxGene), ENCODE, BLUEPRINT
Evaluated all three scientific angles (generic, M1/M2, disease-specific)
Assessed single-cell vs. bulk data availability
Identified 9 high-quality datasets across 3 tiers
Confirmed all major omics layers are available

Decisions locked

Scientific angle: Generic mono→mac + TAM extension
Route: Hybrid sc (primary) + bulk (epigenomic anchor)
Design: 3-tier multi-dataset (discovery → validation → extension)
Positive controls: PU.1, CEBPA, CEBPB, IRF8, MAFB

✅ Phase complete. All strategic decisions made. Dataset stack locked.

Data Acquisition

Download and verify all datasets from public repositories

Completed ▶

Acquisition complete ✓

9 datasets acquired, structurally verified, and organized in a 3-tier design
16 GB total data across transcriptomic, chromatin, methylation, and paired multi-omic layers
Tier 1 discovery, Tier 2 independent validation, and Tier 3 disease/polarization extension are all represented

Key datasets verified

CellxGene 1b350d0a — 1,281,499 cells
CellxGene 1b9d8702 — 329,762 cells
GSE118696 methylation + GSE207308 SHARE-seq
DICE monocyte cohort — 91 donors

Key finding from GSE164498

Dataset includes built-in TF knockdown experiments for IRF1, IRF7, IRF9, ID2 — these are causal validation experiments already in the data. Exceptionally useful for Phase 7.

✅ Complete — all 9 datasets verified and ready for preprocessing.

Preprocessing & Trajectory Mapping

QC, normalization, batch correction, pseudotime trajectory

Active ▶

Single-cell pipeline

Quality control — min genes/cell, MT%, doublet removal
Normalization + log-transform
Batch correction: Harmony or scVI
Trajectory inference: RNA velocity or PAGA
Output: pseudotime-ordered mono→mac trajectory

Bulk pipeline

Alignment: STAR / HISAT2
Quantification: Salmon / featureCounts
Normalization: DESeq2 / edgeR
Time-series ordering (7 time points)
Epigenomic QC for ChIP-seq and ATAC-seq

Tools

Scanpy Seurat Harmony scVI STAR Salmon DESeq2 RNA Velocity PAGA

ML Multi-Omics Integration Model

Novel contribution — fuse transcriptomic + epigenomic layers

Pending ▶

What this model does

Fuses scRNA-seq + bulk RNA-seq + ChIP-seq + ATAC-seq + DNA methylation into a joint representation
Learns the regulatory state of the cell at each differentiation stage
Architecture to be decided based on data structure: multi-modal VAE, GNN, or attention-based fusion
Output: integrated embedding that captures chromatin + expression together

This is the thesis novelty

No published study has applied this exact ML integration to mono→mac differentiation at this scale
Combining sc resolution with bulk epigenomic ground truth is the key methodological advance
The model output feeds directly into GRN reconstruction (Phase 4)

Gene Regulatory Network Reconstruction

Build the TF → target gene network from integrated data

Pending ▶

Methods

pySCENIC — GRN from scRNA-seq co-expression + TF motif enrichment
SCENIC+ — adds scATAC chromatin accessibility layer
ARACNe / GENIE3 — bulk RNA alternative / complement

Output

TF → target gene regulatory network
Regulon definitions (each TF + its targets)
Network at each differentiation stage
Input for TF activity scoring in Phase 5

Tools

pySCENIC SCENIC+ ARACNe GENIE3

Score & Rank Master Regulators

Quantify TF activity across the differentiation trajectory

Pending ▶

Scoring methods

decoupleR — TF activity inference from GRN + target expression
DoRothEA / CollecTRI — curated TF–target regulon database
VIPER — regulon-based TF activity (ARACNE output)

Ranking criteria

Activity differential (monocyte vs. macrophage state)
Regulon size (number of direct targets)
Network centrality (hub score)
Consistency across datasets (Tier 1 + Tier 2)

Positive Control Validation

Verify the pipeline by recovering known master regulators

Pending ▶

Must recover (non-negotiable gate)

PU.1 / SPI1 — master myeloid lineage TF
CEBPA — monocyte to macrophage commitment
CEBPB — macrophage activation and M2 polarization
IRF8 — myeloid differentiation regulator
MAFB — macrophage identity maintenance

Acceptance criterion

All 5 positive controls must appear in the top 20% of the ranked TF list
If any fail → pipeline has a bug → fix before any novel claims are made

In-Silico Causality Testing

Simulate TF knockouts — does removing the TF block differentiation?

Pending ▶

Method

CellOracle — simulate TF knockout in the GRN, predict cell state shift
For each top candidate TF: knock out → does macrophage state collapse?
Causal TFs = those whose loss blocks differentiation or reverts to monocyte state

Built-in validation (GSE164498)

This dataset already has IRF1, IRF7, IRF9, ID2 knockdown experiments
Can compare in-silico predictions to actual knockdown transcriptomes
Rare opportunity to validate CellOracle predictions against real data

Tools

CellOracle GSE164498 KD experiments

Immunotherapy Target Nomination

Connect master regulators to actionable therapeutic targets

Pending ▶

Clinical framing

CAR-T fails for solid tumors & TAM-rich microenvironments
The master TFs controlling macrophage fate are immunotherapy targets
Targeting these TFs could reprogram the tumor immune microenvironment

Mapping axes

Checkpoint pathways (PD-L1, CD47, SIRPα)
Macrophage polarization switches (M1↔M2)
Phagocytosis / ADCC enhancers
Cytokine secretion regulators

Databases used

ChEMBL OpenTargets Pan-cancer TAM atlas GSE154763

Wet-Lab Validation Design

Design the experiment — outsource execution to a collaborating lab

Pending ▶

What will be designed

One or two key validation experiments (e.g., TF knockdown in primary CD14+ monocytes, differentiation assay)
Specify cell source, reagents, readouts, expected result
Full experimental protocol included in thesis

Execution model

Mohammad designs computationally — does not perform wet-lab
Execution outsourced to a collaborating lab or MSc student
Professor-recommended approach; standard for computational PhDs

Writing & Final Defense

Thesis document in Persian + potential publishable paper

Pending ▶

Thesis structure (Persian)

Introduction — monocyte biology, CAR-T context, rationale
Methods — data, ML model, GRN, TF scoring
Results — trajectory, network, ranked TFs, knockouts, targets
Discussion — novel candidates, clinical relevance
Conclusion + proposed validation experiment

Deliverables

Ranked, validated list of candidate master regulators
Network model of monocyte→macrophage transition
In-silico knockout predictions (cell fate shifts)
Immunotherapy target nominations
Proposed wet-lab validation experiment

Dataset inventory — click any dataset for details and sources

Tier 1 — Core Discovery