Datasets:

junidude14
/

PBISC-data

paper-releases/pbisc-diffusion/updates/20261012T130738Z-strictpairs-implementation/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/archive/20261012T130738Z-strictpairs-implementation/snapshot/manuscript/tables/results_realbulk.tex ADDED Viewed

	@@ -0,0 +1,24 @@

+\begin{table}[t]
+\centering
+\footnotesize
+\setlength{\tabcolsep}{4pt}
+\caption{Separately measured bulk transfers the condition effect and not the donor. Cells generated from the clinical bulk of two paired cohorts (1,500 cells per sample, base arm at part 16, drawn read-out) are compared with the same donors' observed single cells. The three arms differ only in the expression vector handed to the conditioner: the assay's bulk (raw), that bulk after a ten-pair gene-wise correction fitted with the donor's batch held out (calibrated), and the donor's own pseudobulk (ceiling). Scored on the genes both assays measure in every sample. Condition effect: Pearson correlation between generated and observed within-donor $\log_2$ fold change, influenza-exposed against mock, over all genes and over the 200 genes with the largest observed change; slope is that of generated on observed. Donor recovery: fraction of samples whose summed generated cells are closest, by correlation, to their own observed pseudobulk among all samples of the cohort; chance is one over the number of samples. The last row applies the same rule to the lupus benchmark, where the model is conditioned on pseudobulk of a cohort resembling its training data.}
+\label{tab:realbulk}
+\begin{tabular}{llrrrrr}
+\toprule
+ & & \multicolumn{3}{c}{Condition effect (Randolph)} & \multicolumn{2}{c}{Donor recovery} \\
+\cmidrule(lr){3-5}\cmidrule(lr){6-7}
+Cohort & Arm & $r$, all genes & $r$, top 200 & slope & top-1 & chance \\
+\midrule
+Randolph, 89 donors $\times$ 2 & raw        & 0.635 & 0.785 & 0.355 & 0.6\%  & 0.56\% \\
+                                & calibrated & 0.644 & 0.789 & 0.439 & 1.7\%  & 0.56\% \\
+                                & pseudobulk & 0.573 & 0.736 & 0.376 & 2.8\%  & 0.56\% \\
+\addlinespace
+ImMicro, 107 donors             & raw        & --- & --- & --- & 0.9\%  & 0.93\% \\
+                                & calibrated & --- & --- & --- & 2.8\%  & 0.93\% \\
+                                & pseudobulk & --- & --- & --- & 3.7\%  & 0.93\% \\
+\addlinespace
+Lupus benchmark, 42 donors      & pseudobulk & --- & --- & --- & 31.0\% & 2.4\% \\
+\bottomrule
+\end{tabular}
+\end{table}

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

+\begin{table}[t]
+\centering
+\small
+\caption{Sampler steps change the effect correlation more than the seed does. One model and one condition set; only the number of deterministic DDIM steps changes ($\eta=0$, so a fixed seed gives every arm the same starting noise per cell). The last row is the spread over five sampling seeds at 50 steps, which is the resolution of a single score. Differential-gene overlap and composition are flat from 25 steps upward; $r$ falls by more than two seed standard deviations at 200 steps.}
+\label{tab:sampling}
+\begin{tabular}{lrrr}
+\toprule
+Steps & $r$ & DEG & comp.\ $L_1$ \\
+\midrule
+10  & 0.263 & 58 & 0.237 \\
+25  & 0.258 & 53 & 0.206 \\
+50  & 0.258 & 57 & 0.206 \\
+100 & 0.253 & 63 & 0.205 \\
+200 & 0.240 & 58 & 0.205 \\
+\midrule
+Seed s.d.\ at 50 steps ($n=5$) & 0.005 & 4.7 & 0.002 \\
+\bottomrule
+\end{tabular}
+\end{table}

	@@ -0,0 +1,18 @@

+\begin{table}[t]
+\centering
+\small
+\caption{Results on shared genes in the childhood SLE pseudobulk test. Disease-effect scores are equal-weight means over T cells, B cells, monocytes and NK cells. RMSE measures errors in disease effects. F1 combines coverage and precision in the shared cell-expression space. TVD measures errors in cell-type shares. The first three rows use one completed schedule part. The last two rows add training only for the 2K model. Higher values are better except for RMSE and TVD.}
+\label{tab:interim_results}
+\begin{tabular}{lrrrrrr}
+\toprule
+Checkpoint & Effect $r$ & RMSE & Coverage & Precision & F1 & TVD \\
+\midrule
+2K E1 & 0.436 & 0.1216 & 0.918 & 0.765 & 0.835 & 0.105 \\
+6K E1 & 0.347 & 0.1270 & 0.798 & 0.926 & 0.857 & 0.125 \\
+11.6K E1 & 0.225 & 0.1340 & 0.636 & 0.944 & 0.760 & 0.139 \\
+\midrule
+2K E2 & 0.573 & 0.1126 & 0.941 & 0.707 & 0.807 & 0.115 \\
+2K E3 & 0.689 & 0.1017 & 0.958 & 0.693 & 0.804 & 0.100 \\
+\bottomrule
+\end{tabular}
+\end{table}

	@@ -0,0 +1,24 @@

+\begin{table}[t]
+\centering
+\small
+\caption{Sweeping the support threshold on one checkpoint. Base arm, part 16, one condition set, one sampler seed and 50 deterministic steps; only the support threshold $\tau$ changes, which decides at the last step whether a gene is expressed ($p>\mathrm{sigmoid}(\tau)$). Weights and predicted values are untouched. Zero fraction and detected genes per cell (median) are measured on the same 2,980-gene panel and the same cells as the scores. Coverage is the fraction of observed cells with a generated neighbour inside their fifteenth-neighbour radius. The threshold that matches the observed mean detection rate is $-1.55$; $-2.0$ maximizes $r$ within twice the seed resolution of $-3.0$ while producing 13\% more detected genes than observed.}
+\label{tab:threshold}
+\begin{tabular}{rrrrrrr}
+\toprule
+$\tau$ & zero fraction & genes/cell & $r$ & DEG & comp.\ $L_1$ & coverage \\
+\midrule
+$+0.5$ & 0.9394 & 170 & 0.231 & 53 & 0.221 & 0.737 \\
+$0.0$ (default) & 0.9360 & 177 & 0.241 & 57 & 0.227 & 0.771 \\
+$-0.5$ & 0.9316 & 188 & 0.253 & 62 & 0.231 & 0.797 \\
+$-1.0$ & 0.9255 & 202 & 0.272 & 63 & 0.233 & 0.814 \\
+$-1.5$ & 0.9168 & 223 & 0.291 & 64 & 0.236 & \textbf{0.821} \\
+$-2.0$ & 0.9040 & 257 & 0.322 & 65 & 0.240 & 0.818 \\
+$-2.5$ & 0.8846 & 310 & 0.316 & 68 & 0.242 & 0.807 \\
+$-3.0$ & 0.8547 & 397 & \textbf{0.327} & \textbf{73} & 0.243 & 0.787 \\
+$-3.5$ & 0.8107 & 530 & 0.317 & 64 & 0.245 & --- \\
+$-4.0$ & 0.7529 & 708 & 0.311 & 67 & 0.246 & --- \\
+\midrule
+Observed cells & 0.9155 & 229 & --- & --- & --- & --- \\
+\bottomrule
+\end{tabular}
+\end{table}