-
Notifications
You must be signed in to change notification settings - Fork 0
/
research.html
114 lines (76 loc) · 7.86 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
---
layout: site-page
mathjax: true
---
<h1>My research and current projects</h1>
<p>Below I list my current projects and a brief description of my role in them. Follow this link for <a href="#publications">my list of peer-reviewed publications</a>.
For additional information, you can visit my personal profiles in
<a href="https://scholar.google.com/citations?user=nXf5nh0AAAAJ">Google
Scholar</a>,
<a href="https://arxiv.org/find/q-bio/1/au:+Mosqueiro_T/0/1/0/all/0/1">arXiv</a>
and <a href="">Research Gate</a>. For a list of side projects I, follow this link.</p>
<p>For my CV, click <a href="http://thmosqueiro.vandroiy.com/CV.pdf">here</a>.</p>
<h2>Learning and information integration in Honey Bees</h2>
<p>
My research interests focus on the multiscale investigation of honey bee behavioral traits, starting from the genetic components, how they modulate brain dynamics and learning, and their impact onto individual and group decision making. This is done with a joint effort by a very interdisciplinary group of PIs led by <a href="https://pinterwollmanlab.eeb.ucla.edu/">Dr. Noa Pinter-Wollman (UCLA)</a>, Dr. Brian Smith (Arizona State University), <a href="http://biocircuits.ucsd.edu/huerta/">Dr. Ramon Huerta (UCSD)</a> and <a href="http://gadau.lab.asu.edu/people/jurgen-gadau-phd/">Dr. Jurgen Gadau (Arizona State University)</a>. Currently, my participation has three fronts: (i) development of neural models that explain how olfaction is processed and learned at the main olfactory pathway in insects (Mushroom Bodies); (ii) data analysis of behavioral and odor conditioning experiments with honey bees; and (iii) create models of foraging activity based on the experiments and data analyses.
</p>
<a href="https://www.youtube.com/watch?v=_hZGlT_luLI">
<img src="/files/projects/video.gif" class='post-img' style="width:60%;" />
</a>
<p>
The above animation is a model developed in close collaboration with Dr. Chelsea Cook (Postdoctoral Researcher at Arizona State University). We have been using models and statistical analyses to answer questions about how honey bees decide to persist in foraging at certain patches of flowers and which learning mechanisms shape these decisions.
</p>
<h2>Temporal integration in large Networks of Spiking Neurons</h2>
<p>
One important part of understand learning of olfaction information is the proper understanding of how noisy signals can be integrated over time and how patterns can be reliably extracted. Even under controlled conditions, chemical signals are inherently noisy and out of equilibrium. The figure below summarizes the architecture of the first place where the olfactory information is processed in the insect brain. The arrows not only indicate the direction through which the signal travels, but also when divergences (orange arrows) or convergences (green arrows) of synaptic connectivity occur. Although this plays a central role in how insects are able to learn and detect patterns, this also induces instability in the signal of the following neural layers (blue neurons).
</p>
<img src="/files/projects/NetworkFig.png" class='post-img' style="width:70%;" />
<p>
In parallel with olfaction and learning, I also conduct research on recording and analyzing chemical sensors (i.e. electronic nose). Electronic noses, which are arrays of chemical sensors sensitive to a spectrum of compounds (e.g., carbon dioxide, methane, etc.), are the best analogue to the olfactory sensory modality of both insects and mammals. Time series recorded by electronic noses can be used to model the response of olfactory receptor cells. Beyond the connection to biology, there are many exciting applications for low-energy devices applied to monitoring of environments such as offices and laboratories. Complex statistical filters (such as online signal de-correlation) can be implemented in boards designed for low consumption of energy (NVIDIA Jetson, Panda Board).
</p>
<img src="/files/projects/enose.png"
alt="Picture of the electronic nose" class='post-img' style="width:60%;" />
<h2>Deep Learning for cell reconstruction</h2>
<p>
In collaboration with the <a href="http://wollman.chem.ucla.edu/">Wollman Lab</a>, we are creating models of Deep Neural Networks capable of reconstructing cells and cell organelles based on simple microscopy images. Although we are still in an early stage, we are working to open our source code and turn this into a service that many labs can apply to their own microscopy studies.
</p>
<img src="/files/projects/projsummary_cellrec.png" class='post-img' style="width:70%;" />
<h2>Optimizing Star Joins using <b>Hadoop</b> Systems</h2>
<p>
In collaboration with <a href="https://github.com/jaquejbrito">Jaqueline Brito</a>,
we are interested in the processing of heavy queries in the Cloud by using
parallel frameworks such as Apache MapReduce and Hadoop Spark. In particular,
<a href="https://www.researchgate.net/publication/299426537_Faster_cloud_Star_Joins_with_reduced_disk_spill_and_network_communication">
we recently studied the processing of Star Joins
</a>,
which are very important
in business intelligence and Online Analytical Processing (OLAP).
The figure has a visual representtion of a star join: a central table $F$ is joined with satellite tables $D_1$, $D_2$, $D_3$ and $D_4$. In OLAP applications, this is a challenge because $F$ is excessively large (at least hundreds of gigabytes). Jaqueline and I were awarded a Microsoft Research Grant in to investigate strategies to optimize the processing of Star Joins.
</p>
<img src="/files/projects/star_joins_summary.png" class='post-img' style="width:80%;" />
<p>
I am particularly interested in problems involving joins of large tables, and how we can push the boundaries of the state of art when solving such operations. We should have a prepring soon about our latest ideas and experiments! :)
</p>
<h2>Notch1 in epithelial cells and calcium signaling</h2>
<p>
In collaboration with Dr. Julia Mack, Dr. Guido Faas and the <a href="https://arispelab.mcdb.ucla.edu/">Arispe Lab</a> at UCLA, we investigated the interperplay between the <a href="https://en.wikipedia.org/wiki/Notch_1">NOTCH1</a> and dynamics of intracellular calcium involved in cell enlongation. Although this project is just getting started, our results were part of a recent manuscript that was accepted on <a href="https://www.nature.com/ncomms/">Nature Communications</a>.
</p>
<img src="/files/projects/notch1_vid_natcomm.gif" class='post-img' style="width:80%;" />
<h2>Communication and dominance contest in electric fish</h2>
<p>
In collaboration with <a href="https://github.com/rtuma">Rafael Tuma</a>,
<a href="http://neurobiofisica.ifsc.usp.br/TeamMembers/teammembers.html">Dr. Reynaldo Pinto</a> and
the <a href="http://neurobiofisica.ifsc.usp.br/">Neurobiofisica Lab</a> at the University of São Paulo,
we are interested in understanding whether and how weakly electric fish (often Gymnotus sp.) communicate
using electric signals. Because these fish are very territorialists,
<a href="https://www.researchgate.net/publication/292982370_Non-parametric_Change_Point_Detection_for_Spike_Trains">
we are proposing novel techniques based on time series segmentation
</a>
to investigate the establishment of a dominance hierarchy between pairs of (conspecific) fish
and the how communication shapes the contest for the dominant role.
</p>
<img src="/files/projects/ElectricFish_segmentation.png" class='post-img' style="width:80%;" />
<h2>Deep Learning for the detection of SNPs</h2>
<p>
In a recent collaboration with <a href="https://sergheimangul.wordpress.com/">Dr. Serghei Mangul</a>, we are starting to look at the possibility of using Deep Learning for SNP detections. Note that Google has recently won the <a href="https://precision.fda.gov/challenges/truth/results">PrecisionFDA Truth Challenge</a> for their highest performance in SNP detection by employing a Deep Neural Network.
</p>