Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261013T034842Z-expanded-131-slide-master/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/expanded_master_20261013/notes.md

@@ -0,0 +1,21 @@
+# Expanded master palette audit
+
+All 66 unique assets in the original 76-slide specification have explicit rules. Existing 32 rules remain unchanged as JSON values; 40 classifications are added, giving 72 rules including six current-review-only assets.
+
+The three legacy rare-state plots use `tools/expanded_legacy_rare.py:build`. This wrapper reroutes output directories and changes only FEW/POOL method-colour tuples. The original local COH cohort palette remains intact, including Ji orange. Regenerated `fig5_rare_states.csv` must equal the original table exactly, including row order and missing values. These are legacy estimates, not revised strict-pair results. The policy supplies `precolored_source_png`, `source_pdf` and `precolored_proof`; select these alternates even when the remaining RGB mapping is empty. No spatial colour-editing extension is needed.
+
+Planning cards and paired-subset readiness panels retain status colours. Corpus and denoiser component colours remain biological/component encodings. Observed rare-state scatters retain cohort colours. Neutral observed and pseudobulk-oracle references follow the existing policy.
+
+The Figure 1 overview retains its existing manuscript vector PDF override. Corpus and planning PNGs do not need method recolouring. No unavailable vector counterpart blocks the policy.
+
+No original data, scripts, estimates, or assets were changed by this audit.
+
+## Inventory conclusion
+
+Healthy/control additions: nineteen distinct panels planned from saved ImMicro native-depth control studies, reference sensitivity, SLE conditioning/split-half controls, technical repeat and annotation QC. No repeated inference or embedding computation. Native depth/output masks use donor sc information; ten-pair calibration and AIFI/pretraining overlap stay explicit.
+
+Disease additions: completed six-unit strict score subset (Sepsis draws0–4, Imperialdraw0), patient-aware DEG, realism controls, influenza/COVID external diagnostics, rare misses/support scaling/cross-disease/C1Q, BehcetK4000, pairedK4000 tradeoffs and historical bulk-only candidate audit. No newdataFigure6completion claim.
+
+Six already-built review views are inserted once; duplicate historical styling copies and superseded views are excluded from additions. Original76slidenumbercoverage willbe explicitlyverified.
+
+Legacy rare figure color rerender reproduces all448saved numeric rows exactly, including roworder andmissingness; scientificzero/donor limitations remainlabelled.

paper-releases/pbisc-diffusion/updates/20261013T034842Z-expanded-131-slide-master/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/expanded_master_20261013/original_76_coverage.csv

@@ -0,0 +1,77 @@
+original_slide,expanded_slide,original_title,current_title
+1,1,PBISC-Diffusion: bulk-to-single-cell generation,PBISC-Diffusion: expanded evidence deck
+2,2,Outline,Full evidence map
+3,4,Manuscript plan: six main figures,Manuscript plan: six main figures
+4,6,Figure 1. Design and performance of PBISC,Figure 1. Design and performance of PBISC
+5,7,Figure 1 plan: Design and performance of PBISC,Figure 1 plan: Design and performance of PBISC
+6,8,Backbone: SDD → PBISC-Diffusion,Backbone: SDD → PBISC-Diffusion
+7,9,Overview draft (Nature style),Overview draft (Nature style)
+8,10,Bulk conditioning,Bulk conditioning
+9,11,Pretraining corpus,Pretraining corpus
+10,12,Width & context ablation: loss saturation,Width & context ablation: loss saturation
+11,13,Width & context ablation: benchmark,Width & context ablation: benchmark
+12,14,Gene context: 2k / 3k / 6k / 15k,Gene context: 2k / 3k / 6k / 15k
+13,16,"PBISC-Diffusion: bulk in, cells out","PBISC-Diffusion: bulk in, cells out"
+14,17,Denoiser: joint value + support diffusion,Denoiser: joint value + support diffusion
+15,19,Figure 2. Fine-tuning PBISC with LoRA,Figure 2. Fine-tuning PBISC with LoRA
+16,20,Figure 2 plan: Fine-tuning PBISC with LoRA,Figure 2 plan: Fine-tuning PBISC with LoRA
+17,21,Real bulk ≠ pseudobulk (model input),Real bulk ≠ pseudobulk (model input)
+18,22,Same shift in three paired cohorts,Same shift in three paired cohorts
+19,23,Consequence: cells off-manifold,Consequence: cells off-manifold
+20,38,Fine-tuning with paired bulk + sc,Fine-tuning with paired bulk + sc
+21,39,Evaluation design,Evaluation design
+22,41,How we score generated cells,How we score generated cells
+23,44,Healthy cohort: LoRA cells cover the real ones,Healthy cohort: LoRA cells cover the real ones
+24,45,Healthy cohort: realism beyond precision/coverage,Healthy cohort: realism beyond precision/coverage
+25,48,Healthy cohort: full vs LoRA vs condition encoder,Healthy cohort: full vs LoRA vs condition encoder
+26,49,Healthy cohort: how many pairs,Healthy cohort: how many pairs
+27,50,Healthy cohort: LoRA rank and K,Healthy cohort: LoRA rank and K
+28,52,Figure 3. PBISC vs scSemiProfiler on healthy donors,Figure 3. PBISC vs scSemiProfiler on healthy donors
+29,53,Figure 3 plan: PBISC vs scSemiProfiler on healthy donors,Figure 3 plan: PBISC vs scSemiProfiler on healthy donors
+30,54,"Healthy donors, 5 pairs: LoRA-5 vs scSemiProfiler-5","Healthy donors, 5 pairs: LoRA-5 vs scSemiProfiler-5"
+31,56,"Healthy donors: the same pairs, 5 to all",Healthy paired-data learning curves
+32,57,Healthy donors: scSemiProfiler copies its representative,Representative-dependent composition
+33,58,"Time: days for scSemiProfiler, minutes for LoRA","Time: days for scSemiProfiler, minutes for LoRA"
+34,59,5 donors: time to apply,5 donors: time to apply
+35,60,Figure 4. Disease signal on paired cohorts,Figure 4. Disease signal on paired cohorts
+36,61,Figure 4 plan: Disease signal on paired cohorts,Figure 4 plan: Disease signal on paired cohorts
+37,62,Five paired cohorts,Five paired cohorts
+38,63,Disease signal with 5 pairs: LoRA-5 vs others,Disease signals with five training donors
+39,67,5-fold CV: disease signal,5-fold CV: disease signal
+40,68,Disease genes at every list length (CAT),Disease genes at every list length (CAT)
+41,71,"Cell-type deconvolution, same pairs: disease signal","Cell-type deconvolution, same pairs: disease signal"
+42,72,SLE benchmark: PBISC vs scSemiProfiler,SLE benchmark: PBISC vs scSemiProfiler
+43,73,SLE benchmark: per cell type,SLE benchmark: per cell type
+44,82,Sepsis: fine-tuning variants,Sepsis: fine-tuning variants
+45,83,Pairs needed: donor and disease signal,Pairs needed: donor and disease signal
+46,84,"Cell-type deconvolution, same pairs: composition","Cell-type deconvolution, same pairs: composition"
+47,86,Figure 5. Rare cell states in disease cohorts,Figure 5. Rare cell states in disease cohorts
+48,87,Figure 5 plan: Rare cell states in disease cohorts,Figure 5 plan: Rare cell states in disease cohorts
+49,88,Rare states with 5 pairs: how many cells,Rare-state fractions: original analysis
+50,90,Rare states with 5 pairs: in the right donors,Rare-state ranking: original analysis
+51,94,5-fold CV: rare states,Rare states: original five-fold analysis
+52,98,Consequence: disease direction flips,Consequence: disease direction flips
+53,102,Figure 6. Discovery from public bulk-only data,Figure 6. Discovery from public bulk-only data
+54,103,Figure 6 plan: Discovery from public bulk-only data,Figure 6 plan: Discovery from public bulk-only data
+55,104,Next: cohorts with a paired subset (planned),Bulk-only cohort plan: original snapshot
+56,108,Supplementary,Supplementary
+57,109,Supplementary plan,Supplementary plan
+58,110,5 donors: LoRA vs scSemiProfiler — cells,5 donors: LoRA vs scSemiProfiler — cells
+59,111,5 donors: realism beyond precision/coverage,5 donors: realism beyond precision/coverage
+60,112,"5 donors: precision/coverage at k = 3, 5, 10","5 donors: precision/coverage at k = 3, 5, 10"
+61,113,5-fold: realism beyond precision/coverage,5-fold: realism beyond precision/coverage
+62,114,Same training pairs: LoRA vs scSemiProfiler,Same training pairs: LoRA vs scSemiProfiler
+63,115,scSemiProfiler copies its representative,scSemiProfiler and the representative-copy baseline
+64,116,Full vs LoRA vs cond.-encoder-only,Full vs LoRA vs cond.-encoder-only
+65,117,"Pairs needed: composition, coverage, training steps","Pairs needed: composition, coverage, training steps"
+66,118,"LoRA rank: r = 4, 16, 64","LoRA rank: r = 4, 16, 64"
+67,119,K diversification: no gain,K diversification: no gain
+68,123,Pairs from other cohorts do not transfer,Pairs from other cohorts do not transfer
+69,124,Longer training on other cohorts: gap stays,Longer training on other cohorts: gap stays
+70,125,"No pretraining: same cohort, from scratch","No pretraining: same cohort, from scratch"
+71,126,"No pretraining: other cohorts, from scratch","No pretraining: other cohorts, from scratch"
+72,127,Rare states: from scratch vs pretrained,Rare states: from scratch vs pretrained
+73,128,Appendix,Appendix
+74,129,Appendix: scSemiProfiler without output threshold,Appendix: scSemiProfiler without output threshold
+75,130,Discussion,Discussion
+76,131,Next: other tissues from a few dozen pairs,Next: other tissues from a few dozen pairs

paper-releases/pbisc-diffusion/updates/20261013T034842Z-expanded-131-slide-master/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/expanded_master_20261013/original_source_hashes.json

@@ -0,0 +1,7 @@
+{
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_paper_summary_20260930.pdf": "eeb0d692c6c89cf8389e7b0294862ee54e955204b4a794db3545bafb64779141",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_paper_summary_20260930.pptx": "51bc0f3a23856bd1d71d38b7bfdbe6f75b63e8054515b8b8a8b2106e9a1612e5",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/generate_deck.py": "005bec4082f16a095b2db207ec02fa8155226825ea2e243b406ed0a2108fa180",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/style.py": "0d17caf276ae0e9b25285b8fd8f9b75c55621ad3a36382b705a366f327c89de6",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/slides_spec.json": "6b8fa055e20a3379ce2e50ca19dbbe285d81e4e95d2c42597e236945b8b2bc23"
+}