Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261008T121550Z-deck-fig4-rules-no-broadcast/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/figure_plan_slide_text.json

@@ -0,0 +1,132 @@
+{
+ "section": "figure-plan",
+ "scripts": [
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/figure_plan.py"
+ ],
+ "generated_from": "tools/figure_plan.py (rerun to refresh; do not edit by hand)",
+ "slides": [
+  {
+   "title": "Manuscript plan: six main figures",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_overview.png"
+   ],
+   "keywords": [],
+   "footnote_short": "Nature style: short abstract and introduction, descriptive results, short discussion, detailed methods; up to 17 supplementary figures.",
+   "footnote": "Nature Style Manuscript Writing.pdf (paper-release-20260928/reference-papers): six main figures, up to 17 supplementary. User decisions of 2026-10-06: no PBISC-VAE; scSemiProfiler, a VAE-based method, is the comparison. User decisions of 2026-10-08 (supersede the earlier figure split): Fig. 2 shows no disease metrics; Fig. 3 compares with scSemiProfiler on healthy samples, with time as a main result; Figs 4-6 use the disease cohorts: Fig. 4 disease benchmarks (PBISC versions, scSemiProfiler, and BayesPrism / CIBERSORTx labelled as cell-type deconvolution), Fig. 5 rare-cell fractions and detection, Fig. 6 new findings from the bulk-only sepsis and COVID samples (K = 1,000 and 4,000). Status: Ready = slide exists and fits the panel; Adapt = slide exists; change its scope or style; Draft = new draft exists, not yet approved; Build = results exist; no slide yet; Run = needs a new analysis or experiment; Plan = not planned yet.",
+   "numbers": []
+  },
+  {
+   "title": "Figure 1 plan: Design and performance of PBISC",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_fig1.png"
+   ],
+   "keywords": [],
+   "footnote_short": "Diffusion on each gene's value and detection, conditioned on bulk, writes realistic blood cells. 5 of 6 panels ready.",
+   "footnote": "Panels: a) Backbones compared: scDiffusion, SDD, PBISC-Diffusion [Ready]. b) Overview: pretraining, diffusion, generation [Draft]. c) What the bulk condition contains [Ready]. d) Pretraining corpus: assays, conditions, cell types [Ready]. e) Width and context: lower loss, same biology [Ready]. f) Gene panel: 2k, 3k, 6k, 15k [Ready]. Details in this section: PBISC-Diffusion: bulk in, cells out; Denoiser: joint value + support diffusion.",
+   "numbers": []
+  },
+  {
+   "title": "Figure 2 plan: Fine-tuning PBISC with LoRA",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_fig2.png"
+   ],
+   "keywords": [],
+   "footnote_short": "On a healthy cohort, a few pairs and a LoRA of minutes give cells close to the real ones. No disease metrics. 2 of 9 panels ready.",
+   "footnote": "Panels: a) Real bulk differs from the training input [Ready]. b) Without fine-tuning, cells fall off the real manifold [Ready]. c) LoRA on paired bulk and cells; how it is tested [Adapt; design example: ImMicro instead of sepsis]. d) How generated cells are scored (cells and composition) [Adapt; disease rows move to Fig. 4]. e) Healthy cohort: precision, coverage, composition [Adapt; ImMicro only; drop the disease columns]. f) Healthy cohort: classifier AUROC, MMD, E_N [Adapt; ImMicro only]. g) Full, LoRA and condition-encoder fine-tuning [Adapt; ImMicro only; no logFC]. h) How many pairs are needed [Adapt; ImMicro only; scSemiProfiler diamonds go to Fig. 3]. i) LoRA rank and K choices [Adapt; ImMicro only; supplementary candidate].",
+   "numbers": []
+  },
+  {
+   "title": "Figure 3 plan: PBISC vs scSemiProfiler on healthy donors",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_fig3.png"
+   ],
+   "keywords": [],
+   "footnote_short": "On healthy donors, LoRA cells are closer to the real ones than scSemiProfiler's, in minutes instead of days. 1 of 4 panels ready.",
+   "footnote": "Panels: a) Five donors: generated cells and realism [Adapt; ImMicro only]. b) Same training pairs (5-fold) and equal pair budget (5-40) [Adapt; ImMicro only; add the n = 5-40 budget runs]. c) scSemiProfiler copies its representative [Adapt; ImMicro only]. d) Time: GPU hours and days per experiment [Ready; all cohorts shown: time is a property of the method].",
+   "numbers": []
+  },
+  {
+   "title": "Figure 4 plan: Disease signal on paired cohorts",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_fig4.png"
+   ],
+   "keywords": [],
+   "footnote_short": "LoRA cells recover the paired single cells' disease signal; compared with other PBISC versions, scSemiProfiler and deconvolution. 6 of 6 panels ready.",
+   "footnote": "Panels: a) Five paired cohorts [Ready]. b) With 5 pairs: LoRA-5 vs PBISC versions and scSemiProfiler-5 [Ready; main comparison]. c) 5-fold CV, reported separately [Ready]. d) Disease genes at every list length (CAT) [Ready]. e) vs BayesPrism and CIBERSORTx (cell-type deconvolution) [Ready]. f) Lupus benchmark (sc-pseudobulk input, zero-shot) [Ready]. Details in this section: Sepsis: fine-tuning variants; Pairs needed: donor and disease signal; Cell-type deconvolution, same pairs: composition.",
+   "numbers": []
+  },
+  {
+   "title": "Figure 5 plan: Rare cell states in disease cohorts",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_fig5.png"
+   ],
+   "keywords": [],
+   "footnote_short": "LoRA cells carry rare disease states near their observed fractions, in the donors that have them. 0 of 4 panels ready.",
+   "footnote": "Panels: a) Rare-state fractions, generated vs observed, four paired cohorts [Run; LoRA, scSemiProfiler, zero-shot; BayesPrism states where defined]. b) Finding rare states: per-donor detection [Run; donors with vs without the state]. c) Behçet: C1Q-high monocytes from real bulk [Adapt; add the LoRA-3 result (C1Q-high BD > HC)]. d) Diseases absent from the corpus: Kawasaki MS1, SLE ISG-high T [Build; rare_states_screen_20261004 results].",
+   "numbers": []
+  },
+  {
+   "title": "Figure 6 plan: Discovery from public bulk-only data",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_fig6.png"
+   ],
+   "keywords": [],
+   "footnote_short": "New findings from LoRA cells of bulk-only sepsis and COVID dialysis samples (K = 1,000 and 4,000). 0 of 2 panels ready.",
+   "footnote": "Panels: a) Bulk-only samples generated: sepsis 122, COVID dialysis 164 [Adapt; update: K = 1,000 and 4,000 done]. b) New findings [Run; earlier screen: 0 confirmed beyond the bulk].",
+   "numbers": []
+  },
+  {
+   "title": "Supplementary plan",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figplan_supp.png"
+   ],
+   "keywords": [],
+   "footnote_short": "Bulk deconvolution methods appear here only: PBISC generates single cells, a different task.",
+   "footnote": "Supplementary: Pairs from other cohorts do not transfer; Longer training on other cohorts: gap stays; No pretraining: same cohort, from scratch; No pretraining: other cohorts, from scratch; Rare states: from scratch vs pretrained. Planned: Influenza exposure from raw clinical bulk (Randolph) [Build; manuscript result; placement open]. Appendix: Appendix: scSemiProfiler without output threshold.",
+   "numbers": []
+  },
+  {
+   "title": "Overview draft (Nature style)",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/fig1b_overview_nature.png"
+   ],
+   "keywords": [
+    "Each cell is learned with its own bulk",
+    "Diffusion on each gene's value and detection",
+    "One bulk in, K cells out; optional LoRA"
+   ],
+   "footnote_short": "Draft for Fig. 1b; schematic, no data. Plan: manuscript editorial/fig1_nature_redesign_plan.",
+   "footnote": "Candidate replacement for the engineering diagram (kept in this section as details and for Methods). Source: paper-release-20260928/manuscript/figures/fig01_architecture_nature.svg (scripts/build_figure1_architecture_nature.py).",
+   "numbers": []
+  },
+  {
+   "title": "Pretraining corpus",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/fig1d_corpus.png"
+   ],
+   "keywords": [
+    "41.4 M cells, 22 datasets, 3,199 donors",
+    "88.5% healthy donors; 23.6% cultured in vitro",
+    "Labels describe the corpus; never model input"
+   ],
+   "footnote_short": "Detailed version: Supplementary Fig. S1 of the manuscript.",
+   "footnote": "Source: paper-release-20260928/manuscript/source_data/corpus/ (build_corpus_tables.py) and figures/fig01_corpus_nature.png (build_figure1_corpus.py).",
+   "numbers": [
+    {
+     "value": "41,395,184 cells; 22 datasets; 3,199 donors; healthy 88.5%; in vitro 23.6%",
+     "source": "manuscript/source_data/corpus/corpus_summary.json"
+    }
+   ]
+  }
+ ]
+}

paper-releases/pbisc-diffusion/updates/20261008T121550Z-deck-fig4-rules-no-broadcast/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/sec09_cat_curves_slide_text.json

@@ -0,0 +1,29 @@
+{
+ "section": "09-cat",
+ "scripts": [
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/sec09_cat_curves.py"
+ ],
+ "generated_from": "tools/sec09_cat_curves.py (rerun to refresh; do not edit by hand)",
+ "slides": [
+  {
+   "title": "Disease genes at every list length (CAT)",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/sec09_cat_curves.png"
+   ],
+   "keywords": [
+    "LoRA top model: 3/4 (5 donors), 2/4 (5-fold)",
+    "LoRA-5 above scSemiProfiler-5: 3/4 cohorts (k = 100)",
+    "Chance at k = 100: 0.012–0.016"
+   ],
+   "footnote_short": "Same-sign overlap of observed and generated top-k genes per cell type; no FDR cut-off.",
+   "footnote": "Correspondence at the top (Irizarry et al. 2005): per cell type, every gene tested on both sides is ranked by |limma-voom moderated t| (case vs control, same donor x type pseudobulks as the DEG panels); CAT(k) = share of the observed top-k genes that are also in the generated top-k with the same logFC sign. Mean over T / B / Mono / NK; 5 donors = mean over 5 draws (scSemiProfiler: its 25 draws), the main comparison; 5-fold = pooled call, shown separately. Chance = k / (2G), G = genes tested on both sides (0.012–0.016 at k = 100). No significance threshold, so cohorts where few genes pass FDR (Giroux, Ji) still get a curve. Rank-biased overlap (Webber et al. 2010, p = 0.98) gives the same order: scores_contrast/{fewshot,pooled}_rbo.csv. Giroux: sex-linked genes excluded. Sepsis CAT100 5 donors LoRA 0.092 / scSemi 0.037 / calibration 0.069 / raw 0.028 / chance 0.012; 5-fold LoRA 0.163; Imperial CAT100 5 donors LoRA 0.243 / scSemi 0.133 / calibration 0.198 / raw 0.179 / chance 0.012; 5-fold LoRA 0.270; Giroux CAT100 5 donors LoRA 0.048 / scSemi 0.075 / calibration 0.050 / raw 0.048 / chance 0.016; 5-fold LoRA 0.068; Ji CAT100 5 donors LoRA 0.091 / scSemi 0.030 / calibration 0.021 / raw 0.021 / chance 0.014; 5-fold LoRA 0.152.",
+   "numbers": [
+    {
+     "value": "Sepsis CAT100 5 donors LoRA 0.092 / scSemi 0.037 / calibration 0.069 / raw 0.028 / chance 0.012; 5-fold LoRA 0.163; Imperial CAT100 5 donors LoRA 0.243 / scSemi 0.133 / calibration 0.198 / raw 0.179 / chance 0.012; 5-fold LoRA 0.270; Giroux CAT100 5 donors LoRA 0.048 / scSemi 0.075 / calibration 0.050 / raw 0.048 / chance 0.016; 5-fold LoRA 0.068; Ji CAT100 5 donors LoRA 0.091 / scSemi 0.030 / calibration 0.021 / raw 0.021 / chance 0.014; 5-fold LoRA 0.152",
+     "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scores_contrast/fewshot_cat.csv; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scores_contrast/pooled_cat.csv"
+    }
+   ]
+  }
+ ]
+}

paper-releases/pbisc-diffusion/updates/20261008T121550Z-deck-fig4-rules-no-broadcast/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/secS_deconvolution_slide_text.json

@@ -0,0 +1,53 @@
+{
+ "section": "S-deconvolution",
+ "scripts": [
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/secS_deconvolution.py"
+ ],
+ "generated_from": "tools/secS_deconvolution.py (rerun to refresh; do not edit by hand)",
+ "slides": [
+  {
+   "title": "Cell-type deconvolution, same pairs: composition",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/secS_deconv_composition.png"
+   ],
+   "keywords": [
+    "PBISC closer than BayesPrism in 22/25 draws, than CIBERSORTx in 24/25",
+    "Deconvolution misreads real bulk",
+    "5-fold: PBISC lowest in 4 of 5 cohorts"
+   ],
+   "footnote_short": "Same donors, truth and scorer as PBISC. Reference = the paired donors' cells.",
+   "footnote": "Cell-type deconvolution (cell-type fractions and per-type expression, not single cells) given the same paired donors as PBISC LoRA, on the paired benchmark's own splits, observed truth and scorer: 5-donor draws (reference = the draw's 5 training donors' observed cells, mixture = the test samples' real bulk) and 5-fold CV (reference = each fold's LoRA training + inner-validation pairs; pooled). BayesPrism: recommended gene set (protein-coding minus chrX/Y/M and ribosomal genes, 15-19k), reference states = gate | AIFI L2; its better theta (initial or final) in cell units is shown. CIBERSORTx: official Docker images (x86, run under qemu emulation on server_1), single-cell reference mode with S-mode batch correction (up to 400 reference cells per type), fractions in cell units (divided by each type's mean UMI per reference cell). CIBERSORTx HiRes gives per-sample cell-type expression for the genes it imputes per type (others are written as 1 and count as zero here); it imputed no gene of the four contrast types in 1 Ji ChILI and 2 Giroux draws, which therefore have no CIBERSORTx disease score. Composition L1 over the six-type gate, lower is better. 5-fold: PBISC has the lowest L1 except in Giroux (scSemiProfiler). Pipeline check: on pseudobulk mixtures BayesPrism recovers the observed mRNA shares (e.g. T 0.39 vs 0.38), so the error comes from real bulk. Source: plans/bayesprism_paired_20261006/results/RESULTS_full.md, RESULTS_cibersortx.md.",
+   "numbers": [
+    {
+     "value": "ImMicro: 5 donors L1 PBISC 0.176 / BayesPrism 0.537 / CIBERSORTx 0.491; 5-fold PBISC 0.162 / BayesPrism 0.575 / CIBERSORTx 0.507 / scSemiProfiler 0.217; Sepsis: 5 donors L1 PBISC 0.466 / BayesPrism 0.446 / CIBERSORTx 0.582; 5-fold PBISC 0.362 / BayesPrism 0.475 / CIBERSORTx 0.607 / scSemiProfiler 0.502; Ji ChILI: 5 donors L1 PBISC 0.513 / BayesPrism 0.893 / CIBERSORTx 0.827; 5-fold PBISC 0.504 / BayesPrism 0.883 / CIBERSORTx 0.914 / scSemiProfiler 0.575; Imperial: 5 donors L1 PBISC 0.503 / BayesPrism 0.845 / CIBERSORTx 0.646; 5-fold PBISC 0.376 / BayesPrism 0.887 / CIBERSORTx 0.592 / scSemiProfiler 0.520; Giroux: 5 donors L1 PBISC 0.256 / BayesPrism 1.455 / CIBERSORTx 0.669; 5-fold PBISC 0.258 / BayesPrism 0.878 / CIBERSORTx 0.591 / scSemiProfiler 0.243",
+     "source": "data/secS_deconvolution.csv"
+    }
+   ]
+  },
+  {
+   "title": "Cell-type deconvolution, same pairs: disease signal",
+   "layout": "full",
+   "figures": [
+    "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/secS_deconv_disease.png"
+   ],
+   "keywords": [
+    "logFC r: PBISC ahead in 15/20 draws (BayesPrism), 17/17 (CIBERSORTx)",
+    "Beyond bulk: PBISC ahead in 18/20 and 17/17",
+    "Top-50 DEG mixed: PBISC ahead in 6/20 and 6/17"
+   ],
+   "footnote_short": "5-donor draws, mean over draws. Partial r = cell-type signal beyond the bulk.",
+   "footnote": "Cell-type deconvolution, not single cells. Case-vs-control per-cell-type logFC (mean CP10K; T, B, Monocyte, NK) from BayesPrism's posterior cell-type expression Z and CIBERSORTx HiRes, scored with the benchmark's own functions; partial r = the same r after both logFCs are residualised on the input bulk's logFC (the cell-type signal the bulk does not carry); top-50 DEG = Welch, summed over types (many ties at 0 in Ji and Giroux). 5-donor draws (the 5-pair comparison), mean over draws with a score. Giroux excludes sex-linked genes, as everywhere in the deck. CIBERSORTx: official Docker images (x86, run under qemu emulation on server_1), single-cell reference mode with S-mode batch correction (up to 400 reference cells per type), fractions in cell units (divided by each type's mean UMI per reference cell). CIBERSORTx HiRes gives per-sample cell-type expression for the genes it imputes per type (others are written as 1 and count as zero here); it imputed no gene of the four contrast types in 1 Ji ChILI and 2 Giroux draws, which therefore have no CIBERSORTx disease score. Source: plans/bayesprism_paired_20261006/results/RESULTS_full.md, RESULTS_cibersortx.md.",
+   "numbers": [
+    {
+     "value": "Sepsis: logFC r PBISC 0.350 / BayesPrism 0.241 / CIBERSORTx 0.069; partial r 0.305 / 0.095 / -0.010; DEG 16.4 / 27.0 / 25.4; Ji ChILI: logFC r PBISC 0.333 / BayesPrism 0.193 / CIBERSORTx 0.158; partial r 0.300 / 0.032 / 0.005; DEG 0.2 / 2.0 / 0.0; Imperial: logFC r PBISC 0.749 / BayesPrism 0.598 / CIBERSORTx 0.283; partial r 0.603 / 0.187 / 0.061; DEG 113.6 / 72.4 / 30.8; Giroux: logFC r PBISC 0.297 / BayesPrism 0.371 / CIBERSORTx 0.095; partial r 0.189 / 0.117 / 0.029; DEG 0.2 / 0.2 / 0.0",
+     "source": "data/secS_deconvolution.csv"
+    },
+    {
+     "value": "draws where PBISC LoRA-5 beats BayesPrism: logFC r 15/20, partial r 18/20, top-50 DEG 6 (BayesPrism ahead 8, ties 6); beats CIBERSORTx: logFC r 17/17, partial r 17/17, top-50 DEG 6 (CIBERSORTx ahead 4, ties 7)",
+     "source": "plans/bayesprism_paired_20261006/results/TABLE_fewshot_full.csv, TABLE_fewshot_cibersortx.csv"
+    }
+   ]
+  }
+ ]
+}