-
Notifications
You must be signed in to change notification settings - Fork 0
/
research.html
executable file
·285 lines (262 loc) · 12.8 KB
/
research.html
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
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="description" content="">
<meta name="author" content="">
<link rel="shortcut icon" href="../../assets/ico/favicon.ico">
<title>Sid Jain's research</title>
<!-- Bootstrap core CSS -->
<link href="css/bootstrap.min.css" rel="stylesheet">
<!-- Custom styles for this template -->
<link href="css/jumbotron.css" rel="stylesheet">
<!-- Just for debugging purposes. Don't actually copy this line! -->
<!--[if lt IE 9]><script src="../../assets/js/ie8-responsive-file-warning.js"></script><![endif]-->
<!-- HTML5 shim and Respond.js IE8 support of HTML5 elements and media queries -->
<!--[if lt IE 9]>
<script src="https://oss.maxcdn.com/libs/html5shiv/3.7.0/html5shiv.js"></script>
<script src="https://oss.maxcdn.com/libs/respond.js/1.4.2/respond.min.js"></script>
<![endif]-->
</head>
<body>
<div class="navbar navbar-inverse navbar-fixed-top" role="navigation">
<div class="container">
<div class="navbar-header">
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="index.html">Sid Jain</a>
</div>
<nav class='navbar-collapse collapse'>
<ul class='nav navbar-nav'>
<li>
<a href='research.html'>Research</a>
</li>
<li>
<a href='publications.html'>Publications</a>
</li>
<li>
<a href='teaching.html'>Teaching</a>
</li>
<li>
<a href='contact.html'>Contact</a>
</li>
<li>
<a href='files/sid_cv.pdf'>CV</a>
</li>
</ul>
</nav>
</div>
</div>
<div class="container">
<div class='header-jumbo'>
<h1>Research</h1>
<ul class='list-inline'>
<li>
<a href='#scrnaseq'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Using Neural Networks for Reducing the Dimensions of Single-Cell RNA-Seq Data
</a>
</li>
<li>
<a href='#timepath'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Reconstructing temporal progression of signaling pathways
</a>
</li>
<li>
<a href='#multitask'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Multitask learning of signaling and regulatory networks
</a>
</li>
<li>
<a href='#ipheuristics'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Heuristics for Mixed Integer Programming
</a>
</li>
<li>
<a href='#lnsdialaride'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Large neighborhood search for Dial-a-ride problem
</a>
</li>
<li>
<a href='#clauselearning'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Clause Learning for Constraint Programs
</a>
</li>
<li>
<a href='#solutioncounting'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Solution counting
</a>
</li>
<li>
<a href='#failurerecovery'>
<span class='glyphicon glyphicon-circle-arrow-down'></span>
Automated Failure Recovery
</a>
</li>
</ul>
</div>
<!-- Example row of columns -->
<hr class='featurette-divider'>
<a name='scrnaseq'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Using Neural Networks for Reducing the Dimensions of Single-Cell RNA-Seq Data</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
<p>
While only recently developed, the ability to profile expression
data in single cells (scRNA-Seq) has already led to several
important studies and findings. However, this technology has also
raised several new computational challenges. These include
questions about the best methods for clustering scRNA-Seq data,
how to identify unique group of cells in such experiments, and how
to determine the state or function of specific cells based on
their expression profile. To address these issues we develop and
test a method based on neural networks (NN) for the analysis and
retrieval of single cell RNA-Seq data. We tested various NN
architectures, some of which incorporate prior biological
knowledge, and used these to obtain a reduced dimension
representation of the single cell expression data. We show that
the NN method improves upon prior methods in both, the ability to
correctly group cells in experiments not used in the training and
the ability to correctly infer cell type or state by querying a
database of tens of thousands of single cell profiles. Such
database queries (which can be performed using our web server)
will enable researchers to better characterize cells when
analyzing heterogeneous scRNA-Seq samples. Read more about it <a
href="https://academic.oup.com/nar/article/4056711/Using-neural-networks-for-reducing-the-dimensions">here</a>.
</p>
</div>
</div>
<hr class='featurette-divider'>
<a name='timepath'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Reconstructing temporal progression of signaling pathways</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
<p>
A stimulant (like a virus) to a cell can interact with proteins in
the cell and cause a cascade of signaling pathways that
activate/repress the expression of genes downstream. The
differential expression of such genes can cause further signaling
cascades that cause the differential expression of another set of
genes. While the gene expression at different time points can be
observed, it is not possible to directly observe the signaling
pathways responsible for their differential expression. We
developed a tool, <a
href="http://sb.cs.cmu.edu/timepath">TimePath</a>, to learn the
progression of such signaling pathways across time based on time
series gene expression data and protein-protein interaction data. Read
more about it <a
href="http://bioinformatics.oxfordjournals.org/content/32/12/i253.abstract">here</a>.
We applied the tool to the analysis of HIV-1 dementia which you can read
about <a
href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5389669/">here</a>.
</p>
</div>
</div>
<hr class='featurette-divider'>
<a name='multitask'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Multitask learning of signaling and regulatory networks</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
<p>
We developed a tool, <a href="http://sb.cs.cmu.edu/mtsdrem">MT-SDREM</a>, to jointly learn signaling pathways and regulatory networks using time series gene expression data from multiple related conditions, TF-gene interaction data, and a protein-protein signaling network. Our tool built on an existing tool called SDREM which learnt the signaling pathways and regulatory network for just a single condition. For joint learning, we shared information the following information :- (1) we ensured that the condition-specific networks learnt were consistent – i.e. the edge directions were the same for all networks (2) If a TF was predicted to regulate more than one condition, it’s prior for regulating any condition was increased. Validation with respect to RNAi screen hits, GO analysis and manual examination of the predicted signaling proteins demonstrated the advantage of joint inference. Read more about it <a href="http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1003943">here</a>.
</p>
</div>
</div>
<hr class='featurette-divider'>
<a name='ipheuristics'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Heuristics for Mixed Integer Programming</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
<p>
We programmed a Mixed Integer programming solver from scratch (using existing LP solvers) and experimented with various node selection and branching heuristics with application to the Warehouse Location problem.
</p>
</div>
</div>
<hr class='featurette-divider'>
<a name='lnsdialaride'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Large neighborhood search for Dial-a-ride problem</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
The Dial-a-ride problem involves a set of passengers that have to be picked up and dropped off at various locations within certain time windows. We developed an algorithm for this problem that was able to obtain close to optimal solutions much more quickly than the state of the art algorithms while still being able to prove if the problem was infeasible – something that most other algorithms did not do. Read more about it <a href="papers/2011/darp.pdf">here</a></p>
</div>
</div>
<hr class='featurette-divider'>
<a name='clauselearning'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Clause Learning for Constraint Programs</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
Redundant clause learning is a popular technique used by SAT solvers to speed up search for a solution or proof of infeasibility for a boolean formula. We developed an algorithm to effectively learn clauses for a class of problems called multi-valued satisfiability problems which are a generalization of boolean formulas. Experimental results showed that our algorithm was several times faster than the standard technique of simply converting the multi-valued formula into its boolean equivalent. Read more about it <a href="papers/2010/cmvsat.pdf">here</a> and <a href="papers/2011/cmvsat_aaai11.pdf">here</a>. </p>
</div>
</div>
<hr class='featurette-divider'>
<a name='solutioncounting'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Solution Counting</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
Solution counting for integer programs with only binary variables is a #P-complete problem. The motivation for this project was that several good techniques for lower bounding the number of solutions for such problems had been developed but the only non-trivial upper bounding technique was extremely slow (in fact, in experiments with the benchmarks we were looking at, it always timed out). We developed an upper bounding technique based on dynamic programming that was able to obtain non-trivial upper bounds much faster for several problems. Read more about it <a href="papers/2010/solution_counting.pdf">here</a>
</p>
</div>
</div>
<hr class='featurette-divider'>
<a name='failurerecovery'></a>
<div class="row featurette">
<div class="col-md-8">
<h2 class='featurette-heading'>Automated Failure Recovery</h2>
</div>
</div>
<div class="row">
<div class="col-md-8">
We examined a programming language abstraction called Stabilizers which had been developed for Concurrent ML for automatically recovering from failures in program execution. We wrote a Javascript library for doing the equivalent for web applications (a use case would be if the network connection suddenly died).
</p>
</div>
</div>
<footer>
</footer>
</div> <!-- /container -->
<!-- Bootstrap core JavaScript
================================================== -->
<!-- Placed at the end of the document so the pages load faster -->
<script src="https://ajax.googleapis.com/ajax/libs/jquery/1.11.0/jquery.min.js"></script>
<script src="js/bootstrap.min.js"></script>
</body>
</html>