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/supplementary.tex ADDED Viewed

	@@ -0,0 +1,40 @@

+% Supplementary Information. Compile independently with pdfLaTeX; all dependencies are local.
+% Rewritten from scratch on 2026-10-10 from the paper-summary deck; previous version in archive/pre_rewrite_20261010/.
+\documentclass[pdflatex,sn-nature]{sn-jnl}
+% Keep identical left/right margins on every page for screen reading.
+\geometry{twoside=false,bindingoffset=0pt}
+\usepackage{graphicx}
+\usepackage{amsmath,amssymb}
+\usepackage{booktabs,array,longtable,multirow}
+\usepackage{microtype}
+\usepackage{xcolor}
+\usepackage{adjustbox}
+\input{macros}
+\renewcommand{\theequation}{S\arabic{equation}}
+\begin{document}
+\noindent\textcolor{red}{[HISTORICAL SUPPLEMENTARY DRAFT: legacy analyses have not been reconciled with exact-five-pair support.]}\par\medskip
+\raggedbottom
+\renewcommand\tablename{Supplementary Table}
+\title{Supplementary Information for: Diffusion model with bulk expression conditioning can generate precise single cell dataset from bulk transcriptome with preserved disease specific gene expression signature}
+\input{authors}
+\maketitle
+\raggedbottom
+\section*{Supplementary Notes}
+\renewcommand{\thesection}{Supplementary Note \arabic{section}:}
+\input{sections/supp_notes}
+\section*{Supplementary Methods}
+\input{sections/supp_methods_model}
+\input{sections/supp_methods_eval}
+\section*{Supplementary Tables}
+\input{sections/supp_table_cohorts}
+\input{sections/supp_tables}
+\section*{Supplementary Figures}
+\input{sections/supp_figures}
+\clearpage
+\bibliography{references}
+\end{document}

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

+\begin{table}[t]
+\centering
+\footnotesize
+\setlength{\tabcolsep}{2.2pt}
+\caption{Capacity lowers the loss and does not move the biology; it also removes the read-out's control. Four models differ only in denoiser trunk width and conditioning width, trained for 16 schedule parts on the 6,000-gene panel with the same data, seed, batch size, learning rate and schedule. Scores are means over parts 14--16 on the held-out lupus split under three read-outs of the same checkpoints; $r$ and DEG higher-better, composition $L_1$ lower-better. Validation loss is the denoising objective at part 16. Response is the change in detected genes per cell when the support threshold is lowered by one unit, measured at part 16: the two wide-trunk models barely respond, so their sparsity cannot be set by a threshold at all.}
+\label{tab:architecture}
+\begin{tabular}{lrrrrrrrrrrrrrr}
+\toprule
+ & \multicolumn{2}{c}{Width} & & Val. & \multicolumn{3}{c}{Cut at $p>0.5$} & \multicolumn{3}{c}{Cut at $\tau=-2.0$} & \multicolumn{3}{c}{Drawn from $p$} & Response \\
+\cmidrule(lr){2-3}\cmidrule(lr){6-8}\cmidrule(lr){9-11}\cmidrule(lr){12-14}
+Model & trunk & cond. & Params & loss & $r$ & DEG & $L_1$ & $r$ & DEG & $L_1$ & $r$ & DEG & $L_1$ & genes/unit \\
+\midrule
+Base           & $1\times$ & 512 & 54.2M & 0.1968 & 0.245 & 59.7 & 0.199 & \textbf{0.326} & 65.7 & 0.208 & 0.286 & 67.7 & 0.197 & 77 \\
+Wide condition & $1\times$ & 2,048 & 102.1M & 0.1969 & 0.245 & 61.0 & 0.161 & 0.304 & \textbf{70.0} & \textbf{0.158} & 0.278 & 68.0 & 0.161 & 79 \\
+Wide trunk     & $4\times$ & 512 & 240.0M & \textbf{0.1355} & 0.265 & 60.7 & 0.181 & 0.271 & 64.3 & 0.182 & 0.268 & 61.7 & 0.181 & 10 \\
+Both wide      & $4\times$ & 2,048 & 312.6M & 0.1356 & 0.253 & 58.7 & 0.264 & 0.262 & 62.0 & 0.264 & 0.259 & 59.3 & 0.264 & 6 \\
+\bottomrule
+\end{tabular}
+\end{table}

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

+\begin{table}[t]
+\centering
+\footnotesize
+\setlength{\tabcolsep}{4pt}
+\caption{A subset-level question inside monocytes. Within each sample, C1Q-high-like monocytes are compared with the other monocytes of that sample. Each entry is the module difference as a background-corrected pseudobulk $\log_2$ fold change, followed by the number of samples with a positive difference. Observed cells are GSE198616 (4 Behçet, 4 control); generated cells come from one model conditioned on GSE198533 bulk (9 Behçet, 10 control), at the inherited read-out ($\tau=0$) and at the swept threshold $\tau=-2.0$ (Supplementary Note~S5). [TODO: columns under the drawn read-out.] All three sets use the same frozen annotator, the same 5,762-gene space and the same frozen module definitions. Group sizes differ, so only means, direction consistency and the intervals quoted in the text are compared.}
+\label{tab:behcet}
+\begin{tabular}{llrrrrrr}
+\toprule
+ & & \multicolumn{2}{c}{Observed} & \multicolumn{2}{c}{Generated, $\tau=0$} & \multicolumn{2}{c}{Generated, $\tau=-2.0$} \\
+\cmidrule(lr){3-4}\cmidrule(lr){5-6}\cmidrule(lr){7-8}
+Module & Group & mean & $+$ & mean & $+$ & mean & $+$ \\
+\midrule
+Interferon-$\gamma$ response & Behçet  & $+0.113$ & 3/4 & $+0.317$ & 7/9  & $+0.157$ & 9/9 \\
+                             & Control & $+0.156$ & 3/4 & $+0.234$ & 8/10 & $+0.136$ & 9/10 \\
+\addlinespace
+Pro-inflammatory             & Behçet  & $-0.124$ & 0/4 & $-0.369$ & 0/9  & $-0.343$ & 0/9 \\
+                             & Control & $-0.156$ & 1/4 & $-0.198$ & 1/10 & $-0.371$ & 0/10 \\
+\addlinespace
+Phagocytosis and             & Behçet  & $+0.157$ & 4/4 & $+0.093$ & 6/9  & $+0.011$ & 7/9 \\
+antigen presentation         & Control & $+0.097$ & 4/4 & $+0.121$ & 10/10 & $-0.036$ & 4/10 \\
+\midrule
+\multicolumn{2}{l}{C1Q-high-like share of control monocytes} & \multicolumn{2}{c}{5.00\%} & \multicolumn{2}{c}{0.85\%} & \multicolumn{2}{c}{4.68\%} \\
+\multicolumn{2}{l}{Samples with fewer than 20 such cells} & \multicolumn{2}{c}{1 of 8} & \multicolumn{2}{c}{19 of 19} & \multicolumn{2}{c}{0 of 19} \\
+\bottomrule
+\end{tabular}
+\end{table}

	@@ -0,0 +1,28 @@

+\begin{table}[t]
+\centering
+\footnotesize
+\setlength{\tabcolsep}{4pt}
+\caption{The bulk condition sets the generated disease effect. Held-out lupus split (42 donors, 2,980 shared genes, scANVI six-type gate). Top block: one model (base arm, part 16), one sampler seed and 50 deterministic steps at the inherited read-out; only the bulk that conditions each donor is replaced, so cell $i$ starts from the same noise in every row. Cell $r$ is the median Pearson correlation between matched cells of that row and the true-condition row; ``closest donor'' counts, of 42, whether the generated pseudobulk is nearest to the donor itself or to the donor whose bulk it was given. Two context rows give the scale of a change. Bottom block: absolute scores at the recommended read-out, the semi-profiling baseline on the same donors, genes and annotator, and the resolution of a score. $r$ and DEG are higher-better, composition $L_1$ lower-better.}
+\label{tab:benchmark}
+\begin{tabular}{lrrrrrr}
+\toprule
+ & & & & & \multicolumn{2}{c}{Closest donor} \\
+\cmidrule(lr){6-7}
+Condition given to every donor & $r$ & DEG & comp.\ $L_1$ & cell $r$ & self & source \\
+\midrule
+Donor's own bulk                       & $+0.241$ & 57 & 0.227 & --- & --- & --- \\
+Bulk of the opposite disease group     & $-0.272$ & 16 & 0.430 & 0.972 & 0 & 40 \\
+Bulk of another donor (derangement)    & $+0.140$ & 0  & 0.502 & 0.969 & 0 & 42 \\
+Cohort-mean bulk (null)                & $-0.041$ & 10 & 0.350 & 0.981 & 1 & --- \\
+\textit{Context}: independent noise    & --- & --- & --- & 0.778 & --- & --- \\
+\textit{Context}: 50 vs 200 steps      & --- & --- & --- & 0.996 & --- & --- \\
+\midrule
+\method{}, drawn read-out, part 16     & 0.287 & 73 & 0.225 & --- & --- & --- \\
+\method{}, drawn read-out, parts 14--16 & 0.286 & 67.7 & 0.197 & --- & --- & --- \\
+scSemiProfiler\cite{wang2024scsemiprofiler} & 0.243 & 52 & 0.249 & --- & --- & --- \\
+\midrule
+Resolution: five sampling seeds (s.d.) & 0.005 & 4.7 & 0.002 & --- & --- & --- \\
+Resolution: two training runs (s.d.\ of paired parts) & 0.018 & 7.2 & 0.023 & --- & --- & --- \\
+\bottomrule
+\end{tabular}
+\end{table}