Datasets:

junidude14
/

PBISC-data

paper-releases/pbisc-diffusion/updates/20261013T094815Z-all-celltypes-final-palette-142/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/final_palette_20261013/native/refcal_top_genes.provenance.json ADDED Viewed

	@@ -0,0 +1,26 @@

+{
+  "key": "refcal_top_genes",
+  "source_records": [
+    {
+      "path": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/umap/cell_immicro_5arm/MANIFEST.json",
+      "sha256": "45e6836942e894e51cfba5e4f9925a0d8affa9bbe95f1adebaf87af2830adf1b",
+      "role": "historical input/control provenance"
+    },
+    {
+      "path": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/umap/cell_immicro_5arm/RESULTS.md",
+      "sha256": "83ec5eec39f15ccb7615da81c4b8f6345563b3a224934f3f942d29bf87b9054c",
+      "role": "historical scope and limitations"
+    },
+    {
+      "path": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/umap/cell_immicro_5arm/tables/g_top_genes_refcal.csv",
+      "sha256": "58fea554b40e04ccacd52ac4e960a667456f9b3f7d1e9c00f49b9db89b558eb1",
+      "role": "saved analysis input"
+    }
+  ],
+  "new_training": false,
+  "new_inference": false,
+  "new_embedding": false,
+  "new_statistical_scoring": false,
+  "display_transform_notes": "Existing ranked gene lists only; no DEG testing. Displayed ranks are original saved ranks within direction and label-stable cells. Immediate-early/stress differences may reflect handling or cohort state; that interpretation is untested. Input and output shifts use different normalization and should not be interpreted as an identical quantitative scale.",
+  "palette": "native"
+}

	@@ -0,0 +1,19 @@

+row,model,cell_text,vs_sdd,vs_sdd_basis,source_paths,source_urls,evidence_quotes
+Diffusion space,scDiffusion,128-d latent (SCimilarity AE),,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_scdiffusion_arxiv2401.03968v2.txt,"https://arxiv.org/abs/2401.03968v2 (Bioinformatics 40(9):btae518, 2024)",[scdiff_paper] we used the pre-trained foundation model SCimilarity [29] as the autoencoder || [scdiff_paper] the output is a 128-dimension latent space embedding
+Diffusion space,SDD,Gene space: value + bit,,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt,https://arxiv.org/abs/2502.02448v3,[sdd_paper] our forward–backward process diffuses continuous and discrete variables jointly
+Diffusion space,PBISC-Diffusion,"Value + bit, 6,000 genes",same,SDD paper: continuous + discrete diffused jointly in data space,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/src/pbisc_sdd/diffusion.py ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/src/pbisc_sdd/model.py,,[diffusion_py] Values and active/support variables are diffused jointly in the original gene space. || [model_py] self.state_dim = 2 * config.n_genes
+Denoiser,scDiffusion,"Skip-MLP, predicts noise",,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_scdiffusion_arxiv2401.03968v2.txt,"https://arxiv.org/abs/2401.03968v2 (Bioinformatics 40(9):btae518, 2024)",[scdiff_paper] fully connected layers and a skip-connected structure || [scdiff_paper] as inputs to predict the noise
+Denoiser,SDD,"Skip-MLP, predicts $x_0$",,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_upstream_1d5423c_sparsity_bit_diffusion.py,https://arxiv.org/abs/2502.02448v3 ; https://raw.githubusercontent.com/PhilSid/sparse-data-diffusion/1d5423c13477f2a52269a6f79113b45c6af5510f/sparsity_bit_diffusion.py,"[sdd_paper] we adopt a skip-connected MLP following scDiffusion || [sdd_paper] denoising neural network fθ to predict x || [sdd_upstream] h = self.fc(x) h = h + self.emb_layer(emb) h = self.norm(h) h = self.act(h) || [sdd_upstream] for layer in self.layers: x = layer(x, emb)"
+Denoiser,PBISC-Diffusion,"Same skip-MLP, predicts $x_0$",same,"upstream ResidualBlock h=fc(x)+emb_layer(emb), LayerNorm, SiLU; emb fed to every block; skip adds; PBISC _ConditionedBlock is the same op sequence; heads 2 x Linear(1024,G) = one Linear to 2G",/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/src/pbisc_sdd/model.py,,"[model_py] Skip-connected conditional MLP predicting value x0 and active logits. || [model_py] hidden = self.input_projection(inputs) + self.context_projection(context) hidden = self.normalization(hidden) hidden = self.activation(hidden) || [model_py] for block in self.down_blocks: hidden = block(hidden, context) || [model_py] hidden = block(hidden, context) hidden = hidden + history.pop()"
+Zeros in training,scDiffusion,No explicit zero model,,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt,https://arxiv.org/abs/2502.02448v3,[sdd_paper] lacks explicit sparsity enforcement
+Zeros in training,SDD,Sparsity bit (±1),,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_upstream_1d5423c_sparsity_bit_diffusion.py,https://arxiv.org/abs/2502.02448v3 ; https://raw.githubusercontent.com/PhilSid/sparse-data-diffusion/1d5423c13477f2a52269a6f79113b45c6af5510f/sparsity_bit_diffusion.py,"[sdd_paper] a cross entropy loss (CE on labels in || [sdd_upstream] return F.mse_loss(pred, img), get_classification_error("
+Zeros in training,PBISC-Diffusion,Same bit; | unmeasured masked out,partial,"bit kept (+-1 targets, cut at 0 as in SDD); new = unmeasured genes excluded from loss (THIRD_PARTY_NOTICES: measured-only BCE/L2). SDD bit loss: CE in the paper (Eq. 6) but F.mse_loss in the public code at 1d5423c, so the loss name is left off the slide",/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/src/pbisc_sdd/diffusion.py ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/THIRD_PARTY_NOTICES.md,,"[diffusion_py] A measured active target is represented by ``+support_scale`` and a measured inactive target by ``-support_scale``. || [diffusion_py] Unmeasured coordinates are neutralized to zero and excluded from both loss terms || [notices] adds conditional bulk/depth inputs, a three-state measurement contract, measured-only BCE/L2 losses"
+Zeros at the end,scDiffusion,No zero step (decoder),,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_scdiffusion_arxiv2401.03968v2.txt ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt,"https://arxiv.org/abs/2401.03968v2 (Bioinformatics 40(9):btae518, 2024) ; https://arxiv.org/abs/2502.02448v3",[scdiff_paper] put it into the decoder to get the final gene expression data || [sdd_paper] lacks explicit sparsity enforcement
+Zeros at the end,SDD,Hard cut: bit > 0,,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_upstream_1d5423c_sparsity_bit_diffusion.py,https://arxiv.org/abs/2502.02448v3 ; https://raw.githubusercontent.com/PhilSid/sparse-data-diffusion/1d5423c13477f2a52269a6f79113b45c6af5510f/sparsity_bit_diffusion.py,"[sdd_paper] using the heaviside step function H applied element-wise || [sdd_upstream] (img[:, self.data_channels:] > 0).float()"
+Zeros at the end,PBISC-Diffusion,"Bernoulli, p = σ(logit)",changed,SDD: Heaviside cut on the bit (paper) and (bit > 0) in upstream code; PBISC final read-out draws Bernoulli(sigmoid(logit)); the logit > 0 cut is kept as the inherited default and also reported (methods),/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/scripts/sample_sdd_standalone.py ; /home/sj_server_1/PBISC/paper-release-20260928/manuscript/sections/methods.tex ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/README.md,,[sampler_py] ``bernoulli`` instead draws each gene from its own probability || [sampler_py] probability = torch.sigmoid(sample.active_logits.to(dtype=torch.float32)) || [methods] $p_{ig}=\mathrm{sigmoid}(\ell_{ig})$ || [final_readme] Bernoulli read-out
+Conditioned on,scDiffusion,Cell-type classifier guidance,,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_scdiffusion_arxiv2401.03968v2.txt,"https://arxiv.org/abs/2401.03968v2 (Bioinformatics 40(9):btae518, 2024)",[scdiff_paper] We use the classifier guidance method to perform conditional generation
+Conditioned on,SDD,Nothing (unconditional),,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_upstream_1d5423c_sparsity_bit_diffusion.py ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt,https://raw.githubusercontent.com/PhilSid/sparse-data-diffusion/1d5423c13477f2a52269a6f79113b45c6af5510f/sparsity_bit_diffusion.py ; https://arxiv.org/abs/2502.02448v3,"[sdd_upstream] def forward(self, x, t, x_self_cond=None): || [sdd_paper] We use self-conditioning [2]"
+Conditioned on,PBISC-Diffusion,Bulk + depth + assay masks,changed,"upstream Cell_Unet.forward(x, t, x_self_cond) has no condition input; SDD paper body: every 'condition' match (n=1) is 'self-conditioning'; THIRD_PARTY_NOTICES: PBISC adds conditional bulk/depth inputs",/home/sj_server_1/PBISC/paper-release-20260928/manuscript/sections/introduction.tex ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/THIRD_PARTY_NOTICES.md,,"[intro] It receives a bulk vector, depth features, the fraction of cells whose assay measured each gene and the output assay mask || [notices] adds conditional bulk/depth inputs, a three-state measurement contract, measured-only BCE/L2 losses"
+Generates,scDiffusion,Decoded gene expression,,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_scdiffusion_arxiv2401.03968v2.txt,"https://arxiv.org/abs/2401.03968v2 (Bioinformatics 40(9):btae518, 2024)",[scdiff_paper] The generated embeddings are finally fed into the decoder to obtain full gene expression
+Generates,SDD,Values with exact zeros,,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/refs/sec02_ref_sdd_arxiv2502.02448v3.txt,https://arxiv.org/abs/2502.02448v3,[sdd_paper] we apply an element-wise product to get the sparsified output
+Generates,PBISC-Diffusion,Many cells from one bulk,changed,SDD samples unconditionally; PBISC draws cells_per_bulk cells per bulk condition (separate from recipe size K),/home/sj_server_1/PBISC/paper-release-20260928/manuscript/sections/methods.tex ; /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/scripts/sample_sdd_standalone.py,,"[methods] Cells generated from the same bulk shared its features and output-assay mask. || [sampler_py] cells_per_bulk=external.cells_per_bulk[:keep],"

	@@ -0,0 +1,11 @@

+method,axis,T cell,B cell,Monocyte,NK cell,DC,Other,mean_over_types,sd_of_mean_over_splits
+ceiling_halfA,1006,0.9247084263736448,0.694303490153745,0.7669150880708799,0.6806720501815684,0.6257941417838438,0.3580073595948055,0.675066759359748,0.0050644650436846
+broadcast_disjoint,1006,0.7870498352482388,0.643495294187295,0.6004261434748315,0.5853060405575121,0.5850718279067483,0.357617608426387,0.5931611249668355,0.0041206712105804
+broadcast_full,1006,0.8002627436576976,0.6565523597607414,0.6150986596334688,0.5936139934230875,0.5934070121157953,0.3706091415493232,0.6049239850233524,0.0053270376218546
+diffusion_5seed_mean,1006,0.587595369462987,0.4090506516847655,0.4202878968081549,0.3434535699696265,0.2707881495060845,0.072696349626319,0.3506453311763229,
+scSemiProfiler,1006,0.4645000580402104,0.3117712763310763,0.2628713556711682,0.2302051848053519,0.259618771295669,0.3003679370420825,0.304889097197593,0.0046035539803665
+ceiling_halfA,2279,0.933943936626066,0.6688573815324589,0.7606182951536543,0.6454468064185985,0.5695353914868434,0.2862369069968667,0.644106453035748,0.0049852574482113
+broadcast_disjoint,2279,0.7773240471510383,0.6252921205375295,0.5963395824020391,0.5510333669209025,0.5433565656422912,0.2665507366450229,0.5599827365498039,0.0045922580615235
+broadcast_full,2279,0.7920078916270111,0.6373964274082867,0.6099838892939113,0.5614987294699733,0.5525479933301343,0.2774722266094551,0.5718178596231286,0.0037339641758269
+diffusion_5seed_mean,2279,0.5823708910464249,0.3392914475928396,0.3707861526875181,0.3095091701706704,0.2154459767146311,0.0430277386599951,0.3100718961453466,
+scSemiProfiler,2279,0.4200084949153659,0.2698971428084941,0.2419043341971318,0.1458978186250204,0.238581522350341,0.2184275501806452,0.2557861438461664,0.0036370858418431

	@@ -0,0 +1,21 @@

+{
+  "key": "sle_disjoint_truth_control",
+  "source_records": [
+    {
+      "path": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/sle_broadcast/RESULTS.md",
+      "sha256": "ae1711181961ede39dff846617d7433e2af2d3341a38a9a5d6251c03d8c358b2",
+      "role": "split-half protocol and caveats"
+    },
+    {
+      "path": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/sle_broadcast/tables/splithalf_summary.csv",
+      "sha256": "f21e93e67a4cbf902d4c5d723e709d62f27ac8c4f838f5676a06b9d81d41a87d",
+      "role": "saved analysis input"
+    }
+  ],
+  "new_training": false,
+  "new_inference": false,
+  "new_embedding": false,
+  "new_statistical_scoring": false,
+  "display_transform_notes": "Existing five random half-A/half-B splits of the same 42 held-out SLE donors. Bulk A is formed from cells disjoint from per-type half-B truth. The full-bulk control includes B. The observed type-specific half-A predictor is a split-half reference. Bars and SDs are saved summary values; no split, expression analysis, or hypothesis test is rerun. PBISC values summarize its five existing sampling seeds.",
+  "palette": "native"
+}