-
Notifications
You must be signed in to change notification settings - Fork 0
/
presentation-2.tex
executable file
·208 lines (167 loc) · 7.6 KB
/
presentation-2.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
\begin{filecontents*}{presentation.bib}
@article{Kaelin:2005gm,
author = {Kaelin, William G},
title = {{The concept of synthetic lethality in the context of anticancer therapy.}},
journal = {Nat. Rev. Cancer},
year = {2005}
}
@article{thanei2018xyz,
title={The xyz algorithm for fast interaction search in high-dimensional data},
author={Thanei, Gian-Andrea and Meinshausen, Nicolai and Shah, Rajen D},
journal={The Journal of Machine Learning Research},
volume={19},
number={1},
pages={1343--1384},
year={2018},
publisher={JMLR. org}
}
\end{filecontents*}
\documentclass[8pt]{beamer}
\usetheme{Hannover}
\usepackage[style=numeric,firstinits=true,backend=biber]{biblatex}
\addbibresource{presentation.bib}
\usepackage{todonotes}
\presetkeys{todonotes}{inline}{}
%\geometry{paperwidth=140mm,paperheight=105mm}
\title{Inferring Genetic Interactions: Comparing \texttt{xyz} and \texttt{glinternet}}
\author{Kieran Elmes}
\begin{document}
\begin{frame}[plain]
\maketitle
\end{frame}
\section{Background}
\begin{frame}{The Problem}
\begin{minipage}{0.4\linewidth}
\includegraphics[width=0.8\linewidth]{"cells_2x"}\\[-1ex]
{\tiny Obligatory xkcd\\Source: \href{https://www.xkcd.com/1217/}{https://www.xkcd.com/1217/}}
\end{minipage}%
\begin{minipage}{0.6\linewidth}
\begin{itemize}
\item Killing cancer cells is easy. Killing cancer cells selectively is hard.
\item Synthetic lethal interactions could potentially be used for exactly that.\cite{Kaelin:2005gm}
\item We have approx. 20,000 genes. Experimentally testing all 200 million pairwise interactions is awkward.
\end{itemize}
\end{minipage}
\end{frame}
\begin{frame}{What are we doing}
\begin{itemize}
\item Simulating large(ish) sets of SiRNA knockdowns, and measuring xyz's ability to correctly determine interactions in the data.
\item We can already do this with glinternet, but \texttt{xyz} is (several hundred times?) faster.
\item \texttt{xyz} $= O(p^\alpha)$ for $1 < \alpha < 2$, depending on the strength of the interactions.
\end{itemize}
\end{frame}
\begin{frame}{The Data}
\begin{itemize}
\item SiRNAs can reduce the expression of several hundred genes at once.
\item $X \leftarrow n \times p$ matrix, $n$ trials with $p$ affected genes.
\item $Y \leftarrow p$ fitness values.
\end{itemize}
\end{frame}
%\begin{center}
%\includegraphics[width=\textwidth]{../interactions-perf-manuscript/figures/Figure_Pmat.pdf}
%\includegraphics[width=0.9\textwidth]{figures/Figure_PRFinteractions.pdf}
\begin{frame}{Runtime comparison}
\centering
\includegraphics[width=0.45\linewidth]{time_taken/time_taken_xyzTRUE}%
\includegraphics[width=0.45\linewidth]{time_taken/time_taken_xyzFalse}
\hspace{2.8cm} \texttt{xyz} \hfill \texttt{glinternet} \hspace{1.8cm}
\end{frame}
\begin{frame}{\texttt{glinternet}: $p = 100$}
\includegraphics[width=0.5\linewidth]{output/PrecRecF1_n1000_tno_large0_xyzFalse_}%
\includegraphics[width=0.5\linewidth]{output/PrecRecF1_n1000_tyes_large0_xyzFalse_}
\end{frame}
\begin{frame}
\centering
\includegraphics[width=0.6\linewidth]{output/test_analysis_n1000_large0_xyzFALSE_}
\end{frame}
\begin{frame}{\texttt{xyz}: $p = 100$}
\includegraphics[width=0.5\linewidth]{output/PrecRecF1_n1000_tno_large0_xyzTRUE_}%
\includegraphics[width=0.5\linewidth]{output/PrecRecF1_n1000_tyes_large0_xyzTRUE_}
\end{frame}
\begin{frame}
\centering
\includegraphics[width=0.6\linewidth]{output/test_analysis_n1000_large0_xyzTRUE_}
\end{frame}
\begin{frame}{\texttt{xyz}: $p = 1000$}
\includegraphics[width=0.5\linewidth]{output/PrecRecF1_n10000_tno_large1_xyzTRUE_}%
\includegraphics[width=0.5\linewidth]{output/PrecRecF1_n10000_tyes_large1_xyzTRUE_}
\end{frame}
\begin{frame}
\centering
\includegraphics[width=0.6\linewidth]{PrecRecF1/test_analysis_n10000_large1_xyzTRUE_}
\end{frame}
%\includegraphics[width=\textwidth]{figures/Figure_PRFobservations.pdf}
%\begin{frame}
%\includegraphics[width=0.5\linewidth]{output/NumObservations_n1000_tno_xyzFALSE}%
%\includegraphics[width=0.5\linewidth]{output/NumObservations_n1000_tyes_xyzFALSE}
%\end{frame}
\section{\texttt{xyz} performance}
\begin{frame}{\texttt{xyz}: Number of observations of each pair}
\includegraphics[width=0.5\linewidth]{output/NumObservations_n1000_tno_xyzTRUE}%
\includegraphics[width=0.5\linewidth]{output/NumObservations_n10000_tno_xyzTRUE}%
\end{frame}
\begin{frame}
\includegraphics[width=0.5\linewidth]{"../interactions-perf-manuscript/output/NumObservations_n1000_percGenes"}%
\includegraphics[width=0.5\linewidth]{"../interactions-perf-manuscript/output/NumObservations_n10000_percGenes"}
\end{frame}
\begin{frame}{\texttt{xyz}: Effect strength}
\includegraphics[width=0.5\linewidth]{output/FXstrength_PRF_n1000_tno_mult1_xyzTRUE_}%
\includegraphics[width=0.5\linewidth]{output/FXstrength_PRF_n10000_tno_mult10_xyzTRUE_}
\end{frame}
%\begin{frame}{Effect direction}
%\includegraphics[width=0.5\linewidth]{output/FXstrength_direction_n1000_tno_mult1_xyzTRUE_}%
%\includegraphics[width=0.5\linewidth]{output/FXstrength_direction_n10000_tno_mult10_xyzTRUE_}
%\end{frame}
%\begin{frame}
%\includegraphics[width=0.5\linewidth]{output/FXdiff_n1000_tno_xyzTRUE}%
%\includegraphics[width=0.5\linewidth]{output/FXdiff_n10000_tno_xyzTRUE}
%\end{frame}
%\begin{frame}
%\includegraphics[width=0.5\linewidth]{output/PrecRecF1_n10000_tno_large0_lethalTRUE_lethal}%
%\includegraphics[width=0.5\linewidth]{small_xyz_50times/large_lethal_n5000_tno_large0_xyzTRUE_lethal}
%\end{frame}
\begin{frame}{Scalability: xyz}
\includegraphics[width=0.5\linewidth]{../interactions-perf-manuscript/output/large_lethal_n5000_tno_large0_xyzTRUE_lethal}%
\includegraphics[width=0.5\linewidth]{../interactions-perf-manuscript/output/large_lethal_n5000_tyes_large0_xyzTRUE_lethal}
\end{frame}
\section{Scalability}
\begin{frame}{Scalability: glinternet}
\includegraphics[width=0.5\linewidth]{../interactions-perf-manuscript/output/large_lethal_n5000_tno_large0_xyzFALSE_lethal}%
\includegraphics[width=0.5\linewidth]{../interactions-perf-manuscript/output/large_lethal_n5000_tyes_large0_xyzFALSE_lethal}
\end{frame}
\section{glinternet performance}
\begin{frame}{\texttt{glinternet}: Number of observations of each pair}
\includegraphics[width=0.5\linewidth]{output/NumObservations_n1000_tno_xyzFALSE}%
\includegraphics[width=0.5\linewidth]{output/NumObservations_n1000_tyes_xyzFALSE}%
\end{frame}
\begin{frame}{\texttt{glinternet}: Effect strength}
\includegraphics[width=0.5\linewidth]{output/FXstrength_PRF_n1000_tno_mult1_xyzFALSE_}%
\includegraphics[width=0.5\linewidth]{output/FXstrength_PRF_n1000_tyes_mult1_xyzFALSE_}
\end{frame}
\begin{frame}{\texttt{glinternet}: Effect direction}
\includegraphics[width=0.5\linewidth]{output/FXstrength_direction_n1000_tno_mult1_xyzFALSE_}%
\includegraphics[width=0.5\linewidth]{output/FXstrength_direction_n1000_tyes_mult1_xyzFALSE_}
\end{frame}
\begin{frame}
\includegraphics[width=0.5\linewidth]{output/FXdiff_n1000_tno_xyzFALSE}%
\includegraphics[width=0.5\linewidth]{output/FXdiff_n1000_tyes_xyzFALSE}
\end{frame}
\section{Parameter choices \& model Assumptions}
\begin{frame}{Choice of $L$}
\centering
\includegraphics[width=0.45\linewidth]{"../interactions-perf-manuscript/output/l_diff_n10000_SNR5_tno"}%
\includegraphics[width=0.45\linewidth]{"output/quant_analysis_n10000"}
\end{frame}
\begin{frame}{Distribution of \texttt{xyz} failures}
\centering
\includegraphics[width=0.5\linewidth]{../interactions-perf-manuscript/failed_xyz_distribution}
\end{frame}
\begin{frame}{Summary}
\begin{enumerate}
\item \texttt{xyz} works for data small enough that we don't need it.
\item \texttt{glinternet} (probably) works for data too large to practically use it.
\end{enumerate}
\end{frame}
%\end{center}
\end{document}
% Things to mention: (also see notes)