From b13c582bbb9619bf5869451e24f6e6dade95c849 Mon Sep 17 00:00:00 2001 From: Justin Bedo Date: Mon, 21 Nov 2022 16:59:38 +1100 Subject: draft PCA/bregman --- .gitignore | 1 + slides.bib | 19 +++++++++++++++ slides.tex | 78 ++++++++++++++++++++++++++++++++++++++++++++++++++------------ 3 files changed, 83 insertions(+), 15 deletions(-) diff --git a/.gitignore b/.gitignore index e06371d..c7f916f 100644 --- a/.gitignore +++ b/.gitignore @@ -6,6 +6,7 @@ slides.bcf slides.blg slides.fdb_latexmk slides.fls +slides.xdv slides.log slides.nav slides.out diff --git a/slides.bib b/slides.bib index 366242b..959e293 100644 --- a/slides.bib +++ b/slides.bib @@ -11,3 +11,22 @@ title = {The Statistical Analysis of Compositional Data}, journal = {Journal of the Royal Statistical Society: Series B (Methodological)} } +@article{collins2001generalization, + title={A generalization of principal components analysis to the exponential family}, + author={Collins, Michael and Dasgupta, Sanjoy and Schapire, Robert E}, + journal={Advances in Neural Information Processing Systems}, + volume={14}, + year={2001} +} +@BOOK{Amari2016-ua, + title = "Information geometry and its applications", + author = "Amari, Shun-Ichi", + publisher = "Springer", + series = "Applied Mathematical Sciences", + edition = 1, + month = feb, + year = 2016, + address = "Tokyo, Japan", + language = "en" +} + diff --git a/slides.tex b/slides.tex index 0eaed66..8bf6c26 100644 --- a/slides.tex +++ b/slides.tex @@ -15,6 +15,12 @@ \newcommand{\R}{\mathbb{R}} \newcommand{\bx}{\mathbf{x}} +\newcommand{\bu}{\mathbf{u}} +\newcommand{\bv}{\mathbf{v}} +\newcommand{\X}{\mathbf{X}} +\newcommand{\V}{\mathbf{V}} +\newcommand{\A}{\mathbf{A}} +\newcommand{\I}{\mathbf{I}} \DeclareMathOperator{\alr}{alr} \DeclareMathOperator{\clr}{clr} @@ -22,6 +28,16 @@ \addbibresource{slides.bib} \AtBeginBibliography{\small} +\usepackage{acronym} +\usepackage{xspace} +\renewcommand*{\acsfont}[1]{\textsc{#1}} +\newacro{dms}{Deep Mutational Scanning} +\newacro{clr}{Centred Log-Ratio} +\newacro{alr}{Additive Log-Ratio} +\newacro{pca}{Principal Component Analysis} +\newcommand{\dms}{\ac{dms} + \xspace} + \author{Justin Bed\H{o}} \title{Exploration of deep mutational scanning data with unsupervised methods} \date{December 13, 2022} @@ -30,9 +46,9 @@ \maketitle - \section{Deep Mutational Scanning (DMS) data} + \section{Deep mutational scanning data} - \begin{frame}{Deep Mutational Scanning (DMS) data} Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. + \begin{frame}{\dms data} Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum. @@ -42,44 +58,76 @@ \begin{frame}{Basics} \begin{definition}[Compositional data] Data \(X \in \R^{n \times d}\) is compositional if rows \(\bx_i\) are in the simplex - \[S^d=\{\,\bx \in \R^d : \forall j,x_j > 0 ; \sum_{j=1}^d x_j = \kappa\,\} \] + \[S^d=\{\,\bx \in \R^d : \forall j,x_j > 0 ; \sum_{j=1}^d x_j = \kappa\,\} \] for constant \(\kappa > 0\). \end{definition} Information is therefore given only by the ratios of components and any composition can be normalised to the standard simplex where \(\kappa = 1\) (c.f., dividing by library size). \end{frame} \begin{frame}{Isomorphisms to Euclidean vector spaces} The simplex forms a \(d-1\) dimensional Euclidean vector space \footfullcite{Aitchison1982}: - \begin{definition}[Additive logratio transform] - \[\alr(\bx)_i = \log \frac{x_i}{x_0} \] + \begin{definition}[\ac{alr}] + \[\alr(\bx)_i = \log \frac{x_i}{x_0} \] \end{definition} - \begin{definition}[Center logratio transform] - \[\clr(\bx)_i = \log \frac{x_i}{\left(\prod_{j=1}^d x_j\right)^{\frac 1 d}} \] + \begin{definition}[\ac{clr}] + \[\clr(\bx)_i = \log \frac{x_i}{\left(\prod_{j=1}^d x_j\right)^{\frac 1 d}} \] \end{definition} \end{frame} - \begin{frame}{PCA on DMS data} + \begin{frame}{\textsc{Pca} on + \ac{dms} data} \begin{block}{Transformation approach} \begin{enumerate} - \item Map DMS data to Euclidean space via ALR/CLR - \item Apply standard PCA + \item Map + \dms data to Euclidean space via + \ac{alr} / + \ac{clr} + \item Apply standard + \ac{pca} \end{enumerate} \end{block} \begin{block}{Problems} \begin{itemize} \item Zeros: \begin{enumerate} - \item geometric mean is \(0\) \(\Rightarrow\) CLR is undefined - \item ALR is undefined for unobserved components + \item geometric mean is \(0\) \(\Rightarrow\) + \ac{clr} is undefined + \item + \ac{alr} is undefined for unobserved components in the ref. \end{enumerate} \item Interpretation: \begin{enumerate} - \item ALR is not isometry - \item CLR is degenerate + \item + \ac{alr} is not isometry + \item + \ac{clr} is degenerate \end{enumerate} \end{itemize} \end{block} \end{frame} - \section{Bregman divergences} + \begin{frame}{Traditional + \ac{pca}} Given \(\X\in \R^{n\times d}\) minimise loss + \[\ell_{\textsc{pca}} \triangleq {\lVert \X - \V\A \rVert}^2_{\textrm{F}} \] + s.t. + \(\V \in \R^{n \times k}\), \(\A \in \R^{k \times d}\), and \(\V^\intercal \V = \I\). + + Has been generalised to exponential families + \footfullcite{collins2001generalization} via Bregman divergences + \footfullcite{Amari2016-ua}. + \end{frame} + + \begin{frame}{Exponential family + \ac{pca}} + \begin{definition}{Bregman Divergence} Let \(\varphi \colon \R^d \to \R\) be a smooth ($C^1$) convex function on convex set \(\Omega\). + The Bregman divergence \(D_\varphi\) with generator \(\varphi\) is + \[ D_\varphi\left(\bu\,\Vert\,\bv\right) \triangleq \varphi(\bu)-\varphi(\bv)-\langle \nabla\varphi(\bv),\bu-\bv\rangle. \] + \end{definition} + + Denote the convex conjugate of \(\varphi\) as \(\varphi^*(\bu) \triangleq \sup_\bv\left\{\langle \bu,\bv\rangle-\varphi(\bv)\right\}\). + The exponential family + \ac{pca} is then given by minimising loss + \[\ell_{\varphi} \triangleq {D_\varphi\left(\X\,\Vert\,\nabla\varphi^*\left(\V\A\right)\right)}^2 \] + under the same constraints as previously, approximating \(\X \sim \nabla\varphi^*\left(\V\A\right)\). + \end{frame} \end{document} -- cgit v1.2.3