Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/strengthening_20261013/rare_literature/ye_observed_marker_summary.csv

@@ -0,0 +1,57 @@
+clinical_group,marker_or_pattern,n_donors,positive_cells,parent_monocytes,mean_donor_percent,median_donor_percent,donors_with_signal
+Pediatric abdominal,CD14,7,7972,9163,77.15343453653747,86.60980810234541,7
+Pediatric abdominal,MKI67,7,783,9163,6.528447352873463,1.02880658436214,5
+Pediatric abdominal,STMN1,7,1204,9163,10.151385358348628,6.172839506172839,6
+Pediatric abdominal,MPO,7,717,9163,5.69144689064759,3.368869936034115,6
+Pediatric abdominal,ELANE,7,671,9163,5.345990500521993,2.857142857142857,6
+Pediatric abdominal,MKI67+STMN1,7,493,9163,4.103247268955951,0.6172839506172839,5
+Pediatric abdominal,MKI67+STMN1 with MPO or ELANE,7,204,9163,1.647816645338764,0.0,3
+Adult other sepsis,CD14,28,31636,37225,80.93467331290071,84.25513987192451,28
+Adult other sepsis,MKI67,28,498,37225,1.1078855327922679,0.8176764869371873,23
+Adult other sepsis,STMN1,28,1360,37225,5.024695875689047,3.7392022359747354,26
+Adult other sepsis,MPO,28,1273,37225,2.5525913078806672,2.6441558441558444,24
+Adult other sepsis,ELANE,28,1184,37225,4.164359451228161,1.8064913946276304,24
+Adult other sepsis,MKI67+STMN1,28,167,37225,0.29748818619277456,0.09873680651160543,17
+Adult other sepsis,MKI67+STMN1 with MPO or ELANE,28,40,37225,0.05604395375361048,0.0,6
+Adult HC,CD14,7,6328,8702,73.67801625480449,75.0,7
+Adult HC,MKI67,7,21,8702,0.2669201163525581,0.18561484918793503,7
+Adult HC,STMN1,7,346,8702,4.184280037174142,4.030418250950571,7
+Adult HC,MPO,7,287,8702,3.4624274848543095,3.108108108108108,7
+Adult HC,ELANE,7,70,8702,1.056913091560118,0.6920415224913494,7
+Adult HC,MKI67+STMN1,7,4,8702,0.058452667571621715,0.07604562737642585,4
+Adult HC,MKI67+STMN1 with MPO or ELANE,7,0,8702,0.0,0.0,0
+Adult ICU,CD14,6,14790,17561,86.57955997884518,85.88262545667527,6
+Adult ICU,MKI67,6,52,17561,0.31287489364293364,0.2949363569500604,6
+Adult ICU,STMN1,6,487,17561,2.6648032760906433,2.839155515042706,6
+Adult ICU,MPO,6,500,17561,2.776816294401757,2.6789408771160264,6
+Adult ICU,ELANE,6,456,17561,3.318244163296584,1.9311434468734163,6
+Adult ICU,MKI67+STMN1,6,12,17561,0.08880694343370406,0.09168697309775498,5
+Adult ICU,MKI67+STMN1 with MPO or ELANE,6,1,17561,0.013082155939298797,0.0,1
+Pediatric HC,CD14,4,2972,4363,65.60319267747286,66.75159520954783,4
+Pediatric HC,MKI67,4,21,4363,0.6219538111234818,0.27744564470465355,4
+Pediatric HC,STMN1,4,142,4363,3.684829392989991,4.105851955691158,4
+Pediatric HC,MPO,4,85,4363,1.710320565896863,2.120541977216959,4
+Pediatric HC,ELANE,4,63,4363,1.2806275648354473,1.3405437845736354,4
+Pediatric HC,MKI67+STMN1,4,8,4363,0.22086569047895013,0.09208103130755065,2
+Pediatric HC,MKI67+STMN1 with MPO or ELANE,4,0,4363,0.0,0.0,0
+Pediatric ICU,CD14,4,5750,8082,74.54208433998883,73.10716283107539,4
+Pediatric ICU,MKI67,4,108,8082,1.025305425932038,1.0340748206785069,4
+Pediatric ICU,STMN1,4,324,8082,4.244225729602302,4.750099676894351,4
+Pediatric ICU,MPO,4,189,8082,2.2761808618826795,2.2810104077928814,4
+Pediatric ICU,ELANE,4,256,8082,2.018635470428967,1.5591200380958075,4
+Pediatric ICU,MKI67+STMN1,4,37,8082,0.38088015967096045,0.3812450570064686,4
+Pediatric ICU,MKI67+STMN1 with MPO or ELANE,4,2,8082,0.010141987829614604,0.0,1
+Pediatric pulmonary,CD14,8,5520,6267,80.51926355845269,87.33325493932104,8
+Pediatric pulmonary,MKI67,8,221,6267,2.931686244954268,1.112945618014515,6
+Pediatric pulmonary,STMN1,8,455,6267,5.528549090464423,3.5328138685029145,6
+Pediatric pulmonary,MPO,8,461,6267,4.635300139381285,4.289418278228322,7
+Pediatric pulmonary,ELANE,8,239,6267,5.162496330444725,4.547449355756065,7
+Pediatric pulmonary,MKI67+STMN1,8,134,6267,2.331039105701644,0.5260234233956208,5
+Pediatric pulmonary,MKI67+STMN1 with MPO or ELANE,8,51,6267,0.4603524798170946,0.01963093835885355,4
+Adult pulmonary,CD14,18,17357,19906,86.852274176148,88.3913988074305,18
+Adult pulmonary,MKI67,18,240,19906,1.2596886861792067,0.9711322545547145,17
+Adult pulmonary,STMN1,18,587,19906,3.5308027932341397,2.968282374975322,18
+Adult pulmonary,MPO,18,938,19906,4.299617538068872,3.128046798830029,17
+Adult pulmonary,ELANE,18,965,19906,4.456600962733879,3.1242287124640065,18
+Adult pulmonary,MKI67+STMN1,18,69,19906,0.3911531668946132,0.20610967979389033,11
+Adult pulmonary,MKI67+STMN1 with MPO or ELANE,18,17,19906,0.10780691471881813,0.0,6

paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/strengthening_20261013/render.log

@@ -0,0 +1,313 @@
+ 7  right fig  9.20x4.86 in  Experimental overview
+ 8  none  fig 12.45x5.67 in  Corpus composition: health, disease risk and cell types
+ 9  none  fig 12.45x5.67 in  From 41.40M source cells to 33.14M training cells
+10  none  fig 12.45x5.67 in  Recipe K and training exposure are different counts
+11  below fig 12.38x5.33 in  Corpus scale and generation workflow
+12  none  fig 12.45x4.07 in  From Gaussian noise to generated cell states
+15  none  fig 12.45x5.67 in  Where context width and trunk width change
+16  none  fig 12.45x5.67 in  Predicted and oracle hints: one trained denoiser
+17  none  fig 12.45x5.67 in  Architecture: full and late validation progression
+18  below fig  9.57x4.63 in  Width & context ablation: benchmark
+19  none  fig 12.45x5.67 in  Architecture arms: dimensions and endpoints
+20  below fig  9.57x4.63 in  Gene context: 2k / 3k / 6k / 15k
+21  none  fig 12.45x5.67 in  Gene panels: progression relative to each endpoint
+22  none  fig 12.45x5.67 in  Gene-panel arms: exact sizes and training provenance
+23  none  fig 12.45x5.67 in  All ablation arms: parameters and training time
+24  none  fig 12.15x5.70 in  Real-cell repeat: a sampling reference
+25  none  fig 12.45x5.48 in  Native-pseudobulk model comparisons
+28  none  fig 11.76x5.70 in  Bulk and pooled cells: different expression values
+29  none  fig 12.45x5.19 in  Paired calibration reduces same-donor expression error
+30  below fig 10.47x4.63 in  Consequence: cells off-manifold
+31  none  fig 12.33x5.70 in  Adaptation recovers the observed cell mixture
+32  none  fig 12.45x5.35 in  Adaptation: mixture error and expression agreement
+33  none  fig 12.15x5.70 in  Healthy paired data: full FT and LoRA
+34  none  fig 12.45x5.40 in  Full FT and LoRA ranks: quality and parameters
+35  none  fig 11.45x5.70 in  What does “all pairs” actually mean?
+36  none  fig 12.19x5.70 in  Healthy pair budget: correlation in all cell types
+37  none  fig 12.19x5.70 in  Healthy pair budget: error in all cell types
+38  right fig  9.20x5.17 in  Five pairs, no extra validation: LoRA vs scSemi
+39  none  fig 12.17x5.70 in  Composition: error and donor-level wins
+42  right fig  9.20x5.18 in  Disease signals with five training donors
+43  right fig  9.20x4.80 in  5-fold CV: disease signal
+44  below fig 11.93x4.63 in  Disease genes at every list length (CAT)
+45  below fig 10.54x4.63 in  Cell-type deconvolution, same pairs: disease signal
+46  none  fig 12.45x5.70 in  Disease gene rankings: complete CV comparison
+47  none  fig 11.99x5.70 in  Disease integrity depends on cohort and endpoint
+48  none  fig 12.19x5.70 in  Five-pair Sepsis: metrics across all support draws
+51  none  fig 12.33x5.70 in  Figure 5 targets begin with the source paper
+title too long (9.93 > 9.78 in): 'Ye sepsis: the published markers in observed paired cells'
+ 7  right fig  9.20x4.86 in  Experimental overview
+ 8  none  fig 12.45x5.67 in  Corpus composition: health, disease risk and cell types
+ 9  none  fig 12.45x5.67 in  From 41.40M source cells to 33.14M training cells
+10  none  fig 12.45x5.67 in  Recipe K and training exposure are different counts
+11  below fig 12.38x5.33 in  Corpus scale and generation workflow
+12  none  fig 12.45x4.07 in  From Gaussian noise to generated cell states
+15  none  fig 12.45x5.67 in  Where context width and trunk width change
+16  none  fig 12.45x5.67 in  Predicted and oracle hints: one trained denoiser
+17  none  fig 12.45x5.67 in  Architecture: full and late validation progression
+18  below fig  9.57x4.63 in  Width & context ablation: benchmark
+19  none  fig 12.45x5.67 in  Architecture arms: dimensions and endpoints
+20  below fig  9.57x4.63 in  Gene context: 2k / 3k / 6k / 15k
+21  none  fig 12.45x5.67 in  Gene panels: progression relative to each endpoint
+22  none  fig 12.45x5.67 in  Gene-panel arms: exact sizes and training provenance
+23  none  fig 12.45x5.67 in  All ablation arms: parameters and training time
+24  none  fig 12.15x5.70 in  Real-cell repeat: a sampling reference
+25  none  fig 12.45x5.48 in  Native-pseudobulk model comparisons
+28  none  fig 11.76x5.70 in  Bulk and pooled cells: different expression values
+29  none  fig 12.45x5.19 in  Paired calibration reduces same-donor expression error
+30  below fig 10.47x4.63 in  Consequence: cells off-manifold
+31  none  fig 12.33x5.70 in  Adaptation recovers the observed cell mixture
+32  none  fig 12.45x5.35 in  Adaptation: mixture error and expression agreement
+33  none  fig 12.15x5.70 in  Healthy paired data: full FT and LoRA
+34  none  fig 12.45x5.40 in  Full FT and LoRA ranks: quality and parameters
+35  none  fig 11.45x5.70 in  What does “all pairs” actually mean?
+36  none  fig 12.19x5.70 in  Healthy pair budget: correlation in all cell types
+37  none  fig 12.19x5.70 in  Healthy pair budget: error in all cell types
+38  right fig  9.20x5.17 in  Five pairs, no extra validation: LoRA vs scSemi
+39  none  fig 12.17x5.70 in  Composition: error and donor-level wins
+42  right fig  9.20x5.18 in  Disease signals with five training donors
+43  right fig  9.20x4.80 in  5-fold CV: disease signal
+44  below fig 11.93x4.63 in  Disease genes at every list length (CAT)
+45  below fig 10.54x4.63 in  Cell-type deconvolution, same pairs: disease signal
+46  none  fig 12.45x5.70 in  Disease gene rankings: complete CV comparison
+47  none  fig 11.99x5.70 in  Disease integrity depends on cohort and endpoint
+48  none  fig 12.19x5.70 in  Five-pair Sepsis: metrics across all support draws
+51  none  fig 12.33x5.70 in  Figure 5 targets begin with the source paper
+52  none  fig 12.18x5.70 in  Ye sepsis: published markers in paired scRNA
+53  none  fig 12.33x5.70 in  Transcriptional programs and rare-population identity
+59  none  fig 12.45x2.62 in  Corpus reference: assay and source composition
+60  below fig 12.45x4.61 in  Backbone: SDD → PBISC-Diffusion
+61  below fig 12.45x4.61 in  Bulk conditioning
+62  below fig 11.68x5.33 in  Denoiser: joint value + support diffusion
+63  below fig 12.45x4.71 in  Fine-tuning with paired bulk + sc
+64  below fig 12.35x4.63 in  Evaluation design
+65  below fig 11.19x4.63 in  How we score generated cells
+66  none  fig 12.45x5.31 in  Does the model follow its bulk conditioning?
+67  below fig 10.30x4.63 in  SLE benchmark: PBISC vs scSemiProfiler
+68  below fig 10.79x4.63 in  SLE benchmark: per cell type
+69  none  fig 12.45x5.38 in  Disjoint-cell controls test the bulk-broadcast baseline
+70  none  fig 12.45x5.60 in  SLE: what is added beyond the bulk broadcast?
+71  below fig 12.45x5.24 in  Same shift in three paired cohorts
+72  none  fig 12.45x5.64 in  Input corrections on one real-reference UMAP
+73  none  fig 12.45x4.86 in  Where generated-cell density differs from real cells
+74  none  fig 12.45x4.89 in  Input calibration changes cell-type composition
+75  none  fig 12.45x5.63 in  Same noise, different input: matched-cell shifts
+76  none  fig 12.45x4.77 in  Matched noise separates cell state from cell identity
+77  none  fig 12.45x5.63 in  Input changes can switch the generated cell label
+78  none  fig 12.45x5.58 in  Nearest real cells: distance and label agreement
+79  none  fig 12.34x5.70 in  Composition agreement does not ensure state agreement
+80  none  fig 12.45x5.47 in  Gene-length effects in the bulk input
+81  none  fig 12.45x5.47 in  Gene-length effects within label-stable cells
+82  none  fig 12.45x5.41 in  Raw and TPM shifts persist within cell labels
+83  none  fig 12.40x5.70 in  Reference calibration leaves specific gene-state shifts
+84  none  fig 12.45x5.45 in  Donor trait axes and donor identification differ
+85  none  fig 12.45x4.61 in  Reference choice changes different quality endpoints
+87  none  fig 12.45x4.72 in  Technical repeats and follow-up are different controls
+88  none  fig 12.45x5.32 in  Annotation agreement and difficult categories
+89  none  fig 12.45x5.40 in  Depth matching changes precision and coverage
+90  none  fig 12.45x4.86 in  Detection and library size are separate diagnostics
+91  right fig  9.20x5.18 in  Healthy cohort: LoRA cells cover the real ones
+92  right fig  9.20x4.79 in  Healthy cohort: realism beyond precision/coverage
+93  right fig  9.20x5.14 in  Healthy cohort: full vs LoRA vs condition encoder
+94  below fig  9.37x4.63 in  Representative-dependent composition
+95  below fig 11.39x4.63 in  Full vs LoRA vs cond.-encoder-only
+96  none  fig  9.32x5.70 in  Pair-budget summary: composition, coverage and steps
+97  right fig  9.20x5.09 in  Healthy cohort: LoRA rank and K
+98  right fig  9.20x5.09 in  Healthy paired-data learning curves
+99  right fig  9.20x4.66 in  Time: days for scSemiProfiler, minutes for LoRA
+100  below fig 10.77x4.63 in  5 donors: time to apply
+101  below fig  9.99x4.63 in  Pairs needed: composition, coverage, training steps
+102  below fig 10.08x4.63 in  LoRA rank: r = 4, 16, 64
+103  right fig  8.31x5.70 in  K diversification: no gain
+104  below fig 12.45x3.82 in  Five paired cohorts
+105  none  fig 12.45x4.70 in  Sepsis: exact-five reconstruction across five draws
+106  none  fig 12.45x5.18 in  Sepsis: realism beyond precision and coverage
+107  none  fig 12.45x4.70 in  Imperial: first completed exact-five comparison
+108  below fig 10.01x4.63 in  5 donors: LoRA vs scSemiProfiler — cells
+109  below fig 10.04x4.63 in  5 donors: realism beyond precision/coverage
+110  below fig 10.02x4.63 in  5 donors: precision/coverage at k = 3, 5, 10
+111  below fig 10.09x4.63 in  5-fold: realism beyond precision/coverage
+112  below fig 11.74x4.63 in  Same training pairs: LoRA vs scSemiProfiler
+113  below fig 11.85x4.63 in  scSemiProfiler and the representative-copy baseline
+114  none  fig 10.46x5.70 in  Sepsis: patient-aware disease-gene agreement
+115  none  fig 10.46x5.70 in  Imperial: patient-aware disease-gene agreement
+116  below fig 10.60x4.63 in  Sepsis: fine-tuning variants
+117  right fig  8.44x5.70 in  Pairs needed: donor and disease signal
+118  below fig 10.53x4.63 in  Cell-type deconvolution, same pairs: composition
+119  none  fig 12.45x4.95 in  Deconvolution: expression coverage limits the comparison
+120  none  fig 12.45x5.08 in  K=4,000 recipe: paired reconstruction quality
+121  none  fig 12.45x5.08 in  K=4,000 recipe: disease signal does not improve uniformly
+122  none  fig 12.45x5.22 in  K=4,000 recipe: calibration gains are selective
+123  none  fig 12.45x4.66 in  Influenza: response order versus response genes
+124  none  fig 10.30x5.70 in  COVID: calibration on a shared real-cell map
+125  none  fig 10.82x5.70 in  COVID monocytes: direction after depth matching
+126  none  fig 11.44x5.70 in  COVID monocytes: direction, magnitude and null
+127  none  fig 11.44x5.70 in  COVID CD14 monocytes: direction, magnitude and null
+128  none  fig 12.45x5.02 in  COVID: biological endpoints give a mixed verdict
+129  none  fig 11.83x5.70 in  Rare-state frequencies in observed cells
+130  none  fig 11.35x5.70 in  Rare-state misses: rules 1–8
+131  none  fig 11.72x5.70 in  Rare-state misses: rules 9–15
+132  none  fig 12.06x5.70 in  Rare-state fidelity: zero-aware reanalysis
+133  none  fig 10.76x5.70 in  Sepsis rare states across paired-data budgets
+134  none  fig 12.45x5.62 in  Sepsis rare-state direction across support budgets
+135  none  fig 11.35x5.70 in  Rare-state screen in DRESS, Kawasaki and SLE
+136  right fig  7.24x5.70 in  Consequence: disease direction flips
+137  none  fig 12.45x5.18 in  C1Q: single-gene detection versus co-detection
+138  none  fig 12.45x4.54 in  Behçet: K=4,000 changes direction more than identity
+139  none  fig 12.45x5.59 in  Behcet: one-gene and multi-gene C1Q rules
+140  below fig  9.46x4.63 in  Pairs from other cohorts do not transfer
+141  below fig  9.44x4.63 in  Longer training on other cohorts: gap stays
+142  below fig 10.61x4.63 in  No pretraining: same cohort, from scratch
+143  below fig 10.53x4.63 in  No pretraining: other cohorts, from scratch
+144  below fig 10.20x4.63 in  Rare states: from scratch vs pretrained
+145  below fig 10.12x4.63 in  Appendix: scSemiProfiler without output threshold
+146  none  fig 12.45x5.28 in  Bulk-only exploration: calibration populations differ
+148  right fig  9.20x5.19 in  Rare-state fractions: original analysis
+149  right fig  9.20x5.01 in  Rare-state ranking: original analysis
+150  below fig  9.23x4.98 in  Rare states: original five-fold analysis
+151  right fig  8.01x5.70 in  Bulk-only cohort plan: original snapshot
+152  none  fig 12.45x4.73 in  Historical bulk-only search: no confirmed new finding
+154  below fig 12.45x4.62 in  Next: other tissues from a few dozen pairs
+Traceback (most recent call last):
+  File "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/build_strengthened_master_20261013.py", line 149, in <module>
+    if __name__=='__main__':main()
+                            ^^^^^^
+  File "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/build_strengthened_master_20261013.py", line 137, in main
+    assert sha(src['path'])==src['sha256'],(family,src['path'])
+           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+AssertionError: ('disease_metrics', '/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure_assignment_20261013/STRENGTHENING_PROPOSAL.md')
+ 7  right fig  9.20x4.86 in  Experimental overview
+ 8  none  fig 12.45x5.67 in  Corpus composition: health, disease risk and cell types
+ 9  none  fig 12.45x5.67 in  From 41.40M source cells to 33.14M training cells
+10  none  fig 12.45x5.67 in  Recipe K and training exposure are different counts
+11  below fig 12.38x5.33 in  Corpus scale and generation workflow
+12  none  fig 12.45x4.07 in  From Gaussian noise to generated cell states
+15  none  fig 12.45x5.67 in  Where context width and trunk width change
+16  none  fig 12.45x5.67 in  Predicted and oracle hints: one trained denoiser
+17  none  fig 12.45x5.67 in  Architecture: full and late validation progression
+18  below fig  9.57x4.63 in  Width & context ablation: benchmark
+19  none  fig 12.45x5.67 in  Architecture arms: dimensions and endpoints
+20  below fig  9.57x4.63 in  Gene context: 2k / 3k / 6k / 15k
+21  none  fig 12.45x5.67 in  Gene panels: progression relative to each endpoint
+22  none  fig 12.45x5.67 in  Gene-panel arms: exact sizes and training provenance
+23  none  fig 12.45x5.67 in  All ablation arms: parameters and training time
+24  none  fig 12.15x5.70 in  Real-cell repeat: a sampling reference
+25  none  fig 12.45x5.48 in  Native-pseudobulk model comparisons
+28  none  fig 11.76x5.70 in  Bulk and pooled cells: different expression values
+29  none  fig 12.45x5.19 in  Paired calibration reduces same-donor expression error
+30  below fig 10.47x4.63 in  Consequence: cells off-manifold
+31  none  fig 12.33x5.70 in  Adaptation recovers the observed cell mixture
+32  none  fig 12.45x5.35 in  Adaptation: mixture error and expression agreement
+33  none  fig 12.15x5.70 in  Healthy paired data: full FT and LoRA
+34  none  fig 12.45x5.40 in  Full FT and LoRA ranks: quality and parameters
+35  none  fig 11.45x5.70 in  What does “all pairs” actually mean?
+36  none  fig 12.19x5.70 in  Healthy pair budget: correlation in all cell types
+37  none  fig 12.19x5.70 in  Healthy pair budget: error in all cell types
+38  right fig  9.20x5.17 in  Five pairs, no extra validation: LoRA vs scSemi
+39  none  fig 12.17x5.70 in  Composition: error and donor-level wins
+42  right fig  9.20x5.18 in  Disease signals with five training donors
+43  right fig  9.20x4.80 in  5-fold CV: disease signal
+44  below fig 11.93x4.63 in  Disease genes at every list length (CAT)
+45  below fig 10.54x4.63 in  Cell-type deconvolution, same pairs: disease signal
+46  none  fig 12.45x5.70 in  Disease gene rankings: complete CV comparison
+47  none  fig 11.99x5.70 in  Disease integrity depends on cohort and endpoint
+48  none  fig 12.19x5.70 in  Five-pair Sepsis: metrics across all support draws
+51  none  fig 12.33x5.70 in  Figure 5 targets begin with the source paper
+52  none  fig 12.18x5.70 in  Ye sepsis: published markers in paired scRNA
+53  none  fig 12.33x5.70 in  Transcriptional programs and rare-population identity
+59  none  fig 12.45x2.62 in  Corpus reference: assay and source composition
+60  below fig 12.45x4.61 in  Backbone: SDD → PBISC-Diffusion
+61  below fig 12.45x4.61 in  Bulk conditioning
+62  below fig 11.68x5.33 in  Denoiser: joint value + support diffusion
+63  below fig 12.45x4.71 in  Fine-tuning with paired bulk + sc
+64  below fig 12.35x4.63 in  Evaluation design
+65  below fig 11.19x4.63 in  How we score generated cells
+66  none  fig 12.45x5.31 in  Does the model follow its bulk conditioning?
+67  below fig 10.30x4.63 in  SLE benchmark: PBISC vs scSemiProfiler
+68  below fig 10.79x4.63 in  SLE benchmark: per cell type
+69  none  fig 12.45x5.38 in  Disjoint-cell controls test the bulk-broadcast baseline
+70  none  fig 12.45x5.60 in  SLE: what is added beyond the bulk broadcast?
+71  below fig 12.45x5.24 in  Same shift in three paired cohorts
+72  none  fig 12.45x5.64 in  Input corrections on one real-reference UMAP
+73  none  fig 12.45x4.86 in  Where generated-cell density differs from real cells
+74  none  fig 12.45x4.89 in  Input calibration changes cell-type composition
+75  none  fig 12.45x5.63 in  Same noise, different input: matched-cell shifts
+76  none  fig 12.45x4.77 in  Matched noise separates cell state from cell identity
+77  none  fig 12.45x5.63 in  Input changes can switch the generated cell label
+78  none  fig 12.45x5.58 in  Nearest real cells: distance and label agreement
+79  none  fig 12.34x5.70 in  Composition agreement does not ensure state agreement
+80  none  fig 12.45x5.47 in  Gene-length effects in the bulk input
+81  none  fig 12.45x5.47 in  Gene-length effects within label-stable cells
+82  none  fig 12.45x5.41 in  Raw and TPM shifts persist within cell labels
+83  none  fig 12.40x5.70 in  Reference calibration leaves specific gene-state shifts
+84  none  fig 12.45x5.45 in  Donor trait axes and donor identification differ
+85  none  fig 12.45x4.61 in  Reference choice changes different quality endpoints
+87  none  fig 12.45x4.72 in  Technical repeats and follow-up are different controls
+88  none  fig 12.45x5.32 in  Annotation agreement and difficult categories
+89  none  fig 12.45x5.40 in  Depth matching changes precision and coverage
+90  none  fig 12.45x4.86 in  Detection and library size are separate diagnostics
+91  right fig  9.20x5.18 in  Healthy cohort: LoRA cells cover the real ones
+92  right fig  9.20x4.79 in  Healthy cohort: realism beyond precision/coverage
+93  right fig  9.20x5.14 in  Healthy cohort: full vs LoRA vs condition encoder
+94  below fig  9.37x4.63 in  Representative-dependent composition
+95  below fig 11.39x4.63 in  Full vs LoRA vs cond.-encoder-only
+96  none  fig  9.32x5.70 in  Pair-budget summary: composition, coverage and steps
+97  right fig  9.20x5.09 in  Healthy cohort: LoRA rank and K
+98  right fig  9.20x5.09 in  Healthy paired-data learning curves
+99  right fig  9.20x4.66 in  Time: days for scSemiProfiler, minutes for LoRA
+100  below fig 10.77x4.63 in  5 donors: time to apply
+101  below fig  9.99x4.63 in  Pairs needed: composition, coverage, training steps
+102  below fig 10.08x4.63 in  LoRA rank: r = 4, 16, 64
+103  right fig  8.31x5.70 in  K diversification: no gain
+104  below fig 12.45x3.82 in  Five paired cohorts
+105  none  fig 12.45x4.70 in  Sepsis: exact-five reconstruction across five draws
+106  none  fig 12.45x5.18 in  Sepsis: realism beyond precision and coverage
+107  none  fig 12.45x4.70 in  Imperial: first completed exact-five comparison
+108  below fig 10.01x4.63 in  5 donors: LoRA vs scSemiProfiler — cells
+109  below fig 10.04x4.63 in  5 donors: realism beyond precision/coverage
+110  below fig 10.02x4.63 in  5 donors: precision/coverage at k = 3, 5, 10
+111  below fig 10.09x4.63 in  5-fold: realism beyond precision/coverage
+112  below fig 11.74x4.63 in  Same training pairs: LoRA vs scSemiProfiler
+113  below fig 11.85x4.63 in  scSemiProfiler and the representative-copy baseline
+114  none  fig 10.46x5.70 in  Sepsis: patient-aware disease-gene agreement
+115  none  fig 10.46x5.70 in  Imperial: patient-aware disease-gene agreement
+116  below fig 10.60x4.63 in  Sepsis: fine-tuning variants
+117  right fig  8.44x5.70 in  Pairs needed: donor and disease signal
+118  below fig 10.53x4.63 in  Cell-type deconvolution, same pairs: composition
+119  none  fig 12.45x4.95 in  Deconvolution: expression coverage limits the comparison
+120  none  fig 12.45x5.08 in  K=4,000 recipe: paired reconstruction quality
+121  none  fig 12.45x5.08 in  K=4,000 recipe: disease signal does not improve uniformly
+122  none  fig 12.45x5.22 in  K=4,000 recipe: calibration gains are selective
+123  none  fig 12.45x4.66 in  Influenza: response order versus response genes
+124  none  fig 10.30x5.70 in  COVID: calibration on a shared real-cell map
+125  none  fig 10.82x5.70 in  COVID monocytes: direction after depth matching
+126  none  fig 11.44x5.70 in  COVID monocytes: direction, magnitude and null
+127  none  fig 11.44x5.70 in  COVID CD14 monocytes: direction, magnitude and null
+128  none  fig 12.45x5.02 in  COVID: biological endpoints give a mixed verdict
+129  none  fig 11.83x5.70 in  Rare-state frequencies in observed cells
+130  none  fig 11.35x5.70 in  Rare-state misses: rules 1–8
+131  none  fig 11.72x5.70 in  Rare-state misses: rules 9–15
+132  none  fig 12.06x5.70 in  Rare-state fidelity: zero-aware reanalysis
+133  none  fig 10.76x5.70 in  Sepsis rare states across paired-data budgets
+134  none  fig 12.45x5.62 in  Sepsis rare-state direction across support budgets
+135  none  fig 11.35x5.70 in  Rare-state screen in DRESS, Kawasaki and SLE
+136  right fig  7.24x5.70 in  Consequence: disease direction flips
+137  none  fig 12.45x5.18 in  C1Q: single-gene detection versus co-detection
+138  none  fig 12.45x4.54 in  Behçet: K=4,000 changes direction more than identity
+139  none  fig 12.45x5.59 in  Behcet: one-gene and multi-gene C1Q rules
+140  below fig  9.46x4.63 in  Pairs from other cohorts do not transfer
+141  below fig  9.44x4.63 in  Longer training on other cohorts: gap stays
+142  below fig 10.61x4.63 in  No pretraining: same cohort, from scratch
+143  below fig 10.53x4.63 in  No pretraining: other cohorts, from scratch
+144  below fig 10.20x4.63 in  Rare states: from scratch vs pretrained
+145  below fig 10.12x4.63 in  Appendix: scSemiProfiler without output threshold
+146  none  fig 12.45x5.28 in  Bulk-only exploration: calibration populations differ
+148  right fig  9.20x5.19 in  Rare-state fractions: original analysis
+149  right fig  9.20x5.01 in  Rare-state ranking: original analysis
+150  below fig  9.23x4.98 in  Rare states: original five-fold analysis
+151  right fig  8.01x5.70 in  Bulk-only cohort plan: original snapshot
+152  none  fig 12.45x4.73 in  Historical bulk-only search: no confirmed new finding
+154  below fig 12.45x4.62 in  Next: other tissues from a few dozen pairs
+{"pages": 154, "newpages": {"corpus": [8, 9, 10], "healthy": [31, 32, 35], "disease_metrics": [46, 47, 48], "rare_literature": [51, 52, 53]}, "groups": {"Introduction": [1, 2, 3, 4], "Figure 1": [5, 6, 7, 8, 9, 10, 11, 12], "Figure 2": [13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25], "Figure 3": [26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39], "Figure 4": [40, 41, 42, 43, 44, 45, 46, 47, 48], "Figure 5": [49, 50, 51, 52, 53], "Figure 6": [54, 55, 56], "Supplementary": [57, 58], "Supp S01": [59, 60, 61, 62, 63, 64, 65], "Supp S02": [66, 67, 68, 69, 70], "Supp S03": [71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85], "Supp S04": [86, 87, 88, 89, 90], "Supp S05": [91, 92, 93, 94, 95], "Supp S06": [96, 97, 98, 99, 100, 101, 102, 103], "Supp S07": [104, 105, 106, 107, 108, 109, 110, 111, 112, 113], "Supp S08": [114, 115, 116, 117, 118, 119, 120, 121, 122], "Supp S09": [123, 124, 125, 126, 127, 128], "Supp S10": [129, 130, 131, 132, 133, 134], "Supp S11": [135, 136, 137, 138, 139], "Supp S12": [140, 141, 142, 143, 144], "Supp S13": [145], "Supp S14": [146], "Historical archive": [147, 148, 149, 150, 151, 152], "Discussion": [153, 154]}}
+convert /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261013_strengthened.pptx as a Impress document -> /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261013_strengthened.pdf using filter : impress_pdf_Export:{"UseTaggedPDF":{"type":"boolean","value":"false"},"PDFUACompliance":{"type":"boolean","value":"false"}}
+{"pass": true, "counts": {"spec": 154, "pptx": 154, "pdf": 154}, "failures": []}

paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/strengthening_20261013/render.py

@@ -0,0 +1,20 @@
+"""CPU-only source build, PDF export and verification."""
+from pathlib import Path
+import os
+import subprocess
+import sys
+if os.environ.get('CUDA_VISIBLE_DEVICES') != '':
+    raise SystemExit('CPU-only bounded unit required')
+A=Path(__file__).resolve().parent
+D=A.parents[1]
+commands=[
+    [sys.executable,str(D/'tools/build_strengthened_master_20261013.py')],
+    ['libreoffice','-env:UserInstallation=file:///tmp/pbisc-strengthened-lo-20261013','--headless','--convert-to',
+     'pdf:impress_pdf_Export:{"UseTaggedPDF":{"type":"boolean","value":"false"},"PDFUACompliance":{"type":"boolean","value":"false"}}',
+     '--outdir',str(D),str(D/'PBISC-Diffusion_expanded_master_20261013_strengthened.pptx')],
+    [sys.executable,str(D/'tools/verify_strengthened_master_20261013.py')],
+]
+with (A/'render.log').open('a') as log:
+    for command in commands:
+        subprocess.run(command,cwd=D,stdout=log,stderr=subprocess.STDOUT,check=True,timeout=600)
+print('Build, PDF and automated verification finished',flush=True)