celldynamics_IL17RD

scDiffeq-based trajectory perturbation analysis of IL17RD in fetal brain radial glia

Overview

This repository contains analysis scripts and results for studying the role of IL17RD (IL-17 receptor D) in fetal brain radial glia (RG) development using stochastic differential equation (SDE)-based trajectory modeling.

We use scDiffeq to simulate cell fate trajectories under in silico gene knockouts, enabling genome-wide perturbation screening without experimental intervention.

Pipeline

Raw scRNA-seq atlas (fetal brain, n~50k cells)
        ↓
  [scVI] Latent embedding (dim=30)
        ↓
  [scDiffeq] SDE trajectory model (t=0 → t=1.0)
        ↓
  Step 6: All-gene KO scan (5,011 HVGs)
          → KO at t0: set gene count → 0, re-encode via scVI → paired L2
        ↓
  Step 7: Top gene trajectory simulation
          → Top 150 bio genes → scDiffeq simulation → L2 at t=1.0
        ↓
  Step 8: Gene program validation
          → KO direction (Δz PC1) × atlas gene expression → Pearson r
        ↓
  Residual L2 analysis
          → Correct for expression confound → expression-independent ranking

Key Results

All-gene KO Landscape (Step 6)

• 5,011 HVG scan; cross-timepoint Spearman r = 0.877 (t70d vs t115d)
• Housekeeping gene filter removes MT-*, RPL*, RPS*, MALAT1 etc.

Top-gene Trajectory Simulation (Step 7)

• Scan L2 vs Simulation L2 Spearman r = 0.97 → encoding shift predicts trajectory divergence
• Top divergent genes: NFIA, NFIB, PTN, NRXN1, SOX11, AUTS2

Expression-corrected Ranking (Residual L2)

• 88-91% of raw L2 variance explained by expression level alone
• After correction: NFIA, NFIB, SOX11 remain top; PAX6 improves (raw 33 → corrected rank 12)
• IL17RD: low raw rank (966) but corrected rank (565) shows modest specific effect

Gene Program Validation (Step 8)

• IL17RD KO direction matches 4/4 known targets (NR2F1, FGFR3 upregulated)
• Best known-target match rate among all 6 reference genes

Reference genes

Gene	Role	t70d bio rank	t70d corrected rank
IL17RD	Study target	966	565
PAX6	RG identity TF	33	12
HES1	Notch effector	23	88
ASCL1	Neurogenic TF	451	4441
NEUROG2	Neurogenic TF	481	160
DLX2	GABAergic TF	1337	4079

Scripts

Script	Environment	Description
scripts/step6_allgene_ko_scan.py	scArches_env	Genome-wide KO encoding scan
scripts/step7a_encode_topgenes.py	scArches_env	Re-encode top genes
scripts/step7b_simulate_topgenes.py	scdiffeq_env	Trajectory simulation
scripts/step8_gene_program_validation.py	scArches_env	Gene program validation
scripts/fig_IL17RD_perturbation.R	BrainAtlas (R)	Main perturbation figures (A-D)
scripts/figEF_allgene_landscape.R	BrainAtlas (R)	All-gene landscape figures (E-G)
scripts/residual_l2_ranking.R	BrainAtlas (R)	Expression-corrected ranking (H)
scripts/run_pipeline_step7.sh	bash	Pipeline: step7a + 7b × 2 timepoints

Environments

• scArches_env: scVI, scArches (for encoding/KO perturbation)
• scdiffeq_env: scDiffeq (for trajectory simulation; no scvi import due to JAX conflict)
• BrainAtlas: R 4.3.3 with ggplot2, patchwork, ggrepel

Data

Large binary files (*.h5ad, *.npy, *.ckpt) are not tracked by git.

File	Description
results/trial6/allgene_scan/allgene_ko_scan_t70d_RG_bio.csv	All-gene scan results (t70d)
results/trial6/allgene_scan/allgene_ko_scan_t115d_RG_bio.csv	All-gene scan results (t115d)
results/trial6/allgene_scan/topgene_sim_t70d_RG.csv	Simulation results (t70d)
results/trial6/allgene_scan/topgene_sim_t115d_RG.csv	Simulation results (t115d)
results/trial6/allgene_scan/*_residual.csv	Expression-corrected rankings
results/trial6/gene_expression_recon/gene_program_validation/	Gene program validation

Citation

• scDiffeq: Kaminski et al.
• scVI: Lopez et al., Nature Methods 2018
• scArches: Lotfollahi et al., Nature Biotechnology 2022