Son Pham, ’17

Project: Using Deep Learning to Automatically Learn Feature Representation and Build a Better Classification Model on Protein Sequential Data
Started: Summer 2015
Funding: Bucknell University PUR

ABSTRACT

In theory, deep learning is not new. However, it has recently become one of the most exciting directions that machine learning has witnessed in years. It has had a tremendous impact on image classification. However, there are very few methods that have investigated its use on strictly sequential data, such as those found in biological sequences. This study will aim to investigate the use of deep learning to induce a protein sequence classifier that can outperform existing methods.

ACHIEVEMENTS

  • Poster Presentation – Sigma Xi 2015 Summer Research Symposium
  • Poster Presentation – Fifth Annual Susquehanna Valley Undergraduate Research Symposium, SVURS 2015, August 4, Bucknell University, Lewisburg, PA
  • Poster Presentation – Presented at 15th Annual Kalman Research Symposium, April 2, 2016, Bucknell University, Lewisburg, PA

POST GRADUATION UPDATES

Son graduated with his degrees in Computer Science and Engineering, together with Digital Studio Arts. He went on to work for Amazon as an Software Engineering Intern, then took a position at Google working with machine learning. Son graduated with the aim of going back to graduate school in 1-2 years.

Chuqiao Ren, ’15

Project: A novel ensemble classifier for protein contact map prediction
Duration: Summer 2013 – Spring 2015
Funding: Bucknell University Program for Undergraduate Research, BRK Startup Fund, Geisinger BGRI Grant, CS Dept. Fund

ABSTRACT

One of the greatest challenges in bioinformatics is how to predict the 3-D structure of a protein by understanding the relationship between a sequence and its amino acid structure.  A protein contact map is a useful way of representing protein 3-D conformations. It is based on a distance matrix, which is a symmetric matrix that contains the Euclidean distance between each pair of C-alpha atoms in each residue in the folded protein.  

Our goal is to improve existing machine learning algorithms for predicting a protein contact map from protein sequence, and develop a novel algorithm that improves the performance of existing contact map predictors.

ACHIEVEMENTS

  • Honors Thesis – Successfully defended, April 2015
  • Short paper and poster – ACB BCB ’14 – ACM International Conference on Bioinformatics, Computational Biology and Biomedicine, Sept 20-23, Newport Beach, CA [link]
  • Poster Presentation – Fourth Annual Susquehanna Valley Undergraduate Research Symposium, SVURS 2014, August 5, Geisinger Research, Danville, PA
  • Poster – Kalman Research Symposium 2014, March 29, Bucknell University, Lewisburg, PA.

POST GRADUATION UPDATES

Chuqiao successfully defended her honors thesis in April, 2015. She is staying for a bit longer this summer to help finish a journal publication and submit before she departs us. She is currently planning on pursuing her graduate degree in computer science at Columbia University, starting Fall 2015. Congratulations, Chuqiao!

Summer 2015

We have an active summer in store. Three students are working on entirely different research projects, while Rachel Ren is wrapping up her work.

  • Son Pham is working on investigating the use of Deep Learning for protein sequence classification. Deep Learning has recently gained substantial recognition due to its success with automated image recognition and speech classification. Very few have examined its use in bioinformatics. Son will help me explore this untapped area in bioinformatics.
  • Jason Hammett will be applying data mining techniques to years of regional climate data, including local stats for the Susquehanna River, to develop explanatory and predictive models for anomalistic weather events around the Susquehanna River Valley.
  • Robert Cowen will be continuing the wonderful work that I started with Bucknell Student Stephanie Gonthier last year on word prediction. Robert will be collaborating with myself and speech pathologists at the Geisinger-Bucknell Autism and Developmental Medicine Institute (ADMI) to develop a preliminary version of a new augmentative and alternative communication (AAC) app that will utilize my word prediction model. This first version will be developed to run on Android tablets.
  • Rachel Ren is graciously staying for a month after graduating to help submit a paper based on her extensive work completed for her honors thesis. Stayed tuned!

Spring 2015

Rachel Ren successfully defended her honors thesis, titled, “Predicting Protein Contact Maps by Bagging Decision Trees”. Congratulations, Rachel! Additionally, Rachel will be attending graduate school starting in the fall at Columbia University, where she will pursue a Masters in Computer Science. Rachel intends to focus on research in machine learning.

Congratulations, Rachel! Bucknell is proud of you! We wish you the very best as you pursue your graduate work.

Stephanie Gonthier, ’15

Project: Using statistical learning to improve word prediction for augmentative and alternative communication
Duration: Summer 2014
Funding: Bucknell University Program for Undergraduate Research, Geisinger BGRI Grant

ABSTRACT

There are a multitude of reasons why people may be unable to communicate effectively through verbal speech, including disorders like ALS, MS, Cerebral Palsy and Autism. Some people use augmentative and alternative communication (AAC), which is simply any mode of communication besides verbal speech, including gestures, writing, facial expressions, pointing to pictures and so on. In recent decades, the field of AAC has been flooded by electronic devices which generate speech for these people based on combinations of pictures, symbols and/or words that are stored on the device. Unfortunately, these devices do present problems; notably, the communication rate with a device is reduced to a fraction of the communication rate of normal speakers. The average user of a device is only able to communicate 10 words per minute, compared to the 130-200 words per minute of an average speaker [ref]. This stark contrast can leave users frustrated, reducing the utility of such devices. The aim of this research is to develop a novel algorithm that would increase the communication rate for users of AAC devices.

ACHIEVEMENTS

  • Oral Presentation – Fourth Annual Susquehanna Valley Undergraduate Research Symposium, SVURS 2014, August 5, Geisinger Research, Danville, PA
    Winner for oral presentation – One of three chosen out of 86 submissions!
  • Poster Presentation – 2014 Sigma Xi Summer Student Research Symposium, July 24, Bucknell University, Lewisburg, PA

POST GRADUATION UPDATES

Elizabeth Dwornik, ’14

Project: Named-Entity Recognition
Duration: Summer 2013 – Spring 2014
Funding: Bucknell University Program for Undergraduate Research

ABSTRACT

Liz is working on a system that can annotate all of the named entities within a text. There are good systems that can identify named entities, however, identifying the type of named entity is a more challenging problem. Many successful systems use simple database lookup techniques and identify entities from a master gazetteer. We are working on a system that can distinguish among different types of named entities without a gazetteer. Our initial efforts will focus on distinguishing entities between location, organization, or person. We plan to start by developing a large set of regular expressions that can be used to classify the different types of entities.

ACHIEVEMENTS

  • Poster: Kalman Research Symposium 2013, April 13, Bucknell University, Lewisburg, PA

POST GRADUATION UPDATES

Liz pursued graduate school studies at Carnegie Mellon University, starting Fall 2014. She enrolled in the Software Management program in the Information Networking Institute. Congratulations, Liz!

Matthew Rogge, ’17

Project: Analysis of Spike Timing Dependent Neural Networks for More Efficient Starting State Learning
Duration: Summer 2014
Funding: Bucknell University Program for Undergraduate Research

ABSTRACT

Artifical Neural Networks (ANN) have been a popular machine learning method for decades. They aim to simulate the behavior of the neurons in the biological brain. One particular type of ANN that is an especially accurate representation of biological neurons is the Spike-Timing Dependent ANN. These ANNs differ from traditional back propagation ANNs in that they rely on the timing and frequency of signals, rather than their strength, to learn and process information. This type of ANN of often ignored for many reasons, mostly due to the computational complexity of learning using these models. On substantial challenge lies in the difficulty of determining the initial configuration of the network. The time required to train the network is also a formidable challenge. My research seeks to eliminate one of these hurdles by deriving an efficient algorithm that can determine the proper starting configuration for the ANN.

ACHIEVEMENTS

  • Poster: 2014 Sigma Xi Summer Student Research Symposium, July 24, Bucknell University, Lewisburg, PA

POST GRADUATION UPDATES

Charles Cole ’14

Project: Using Machine Learning to Predict the Health of HIV-Infected Patients
Duration: Summer 2012 – Spring 2014
Funding: Bucknell University PUR, Biology Dept. and CS Dept. Funding

ABSTRACT

HIV is one of the most devastating viruses to hit mankind in modern history. About half of people infected will acquire AIDS. For some, however, the virus will lay in a stage known as “clinical latency” for 10, perhaps up to 20 years; in this stage, the symptoms are mild, sometimes even non-existant. This study aims to investigate the potential existance of specific patterns in the genome of HIV, and the prognosis of the infected patient. Discovery of such patterns could help aid researchers in improved understanding of the genetics of HIV, assisting in identifying potential patterns that researchers should look for to help infected doctors predict patient prognosis more accurately. Moreover, the identification of specific mutations or recurring patterns that are highly deleterious to the infected patient could aid in the development of drugs to target those genes containing the deleterious mutations.

ACHIEVEMENTSS

  • Honors thesis defense passed – April 25, 2014
  • Short paper and poster: ACB BCB ’13 – ACM International Conference on Bioinformatics, Computational Biology and Biomedicine, Sept 22-25, Washington DC
  • Oral presentation: Third Annual Susquehanna Valley Undergraduate Research Symposium, SVURS 2013, August 6, Geisinger Research, Danville, PA
    • Winner for oral presentation – One of three chosen out of 67 submissions!
  • Poster: Kalman Research Symposium 2013, April 13, Bucknell University, Lewisburg, PA.

POST GRADUATION UPDATES

Charles was accepted into to a pre-med program at Temple University, and will be starting medical school immediately thereafter.

Brigitte Hofmeister ’14

Project: Modeling the Evolution of Influenza
Duration: Summer 2013 – Fall 2013
Funding: Bucknell University Program for Undergraduate Research, CS Dept. Funding

OVERVIEW

The primary aim of this project was to develop a model of evolution of the Influenza virus. We were interested in learning if there were any model that could predict future variants better than random. This work was far more complex than we initially envisioned. However, some really interesting

The first project (completed Summer 2013) was to develop an alignment-free model that can assess the similarity between protein sequences. The grand objective, however, was to induce a model of evolution among one or two of the gene products of Influenza. To do this, we started with an n-gram model of the protein, and compute a distance between sequences by not only considering n-grams that are identical, but also those that have high biological similarity. To this end, we incorporate a standard substitution matrix (e.g. BLOSUM62) in the distance calculation between n-grams that do not have a 100% match. This work ended up with our first project, and ultimately the primary outcome that had the most utility: Using n-gram protein models with substitution matrices for phylogenetic analysis.

ABSTRACT

Phylogenetic analyses, specifically phylogenetic tree constructions, are important for understanding evolution and species relatedness. Most methods require a multiple sequence alignment (MSA) to be performed prior to inducing the phylogenetic tree. MSAs, however, are computationally expensive and increasingly error prone as the number of sequences increase, as the average sequence length increases, and as the sequences in the set become more divergent. We introduce a new method called ngPhylo, an n-gram based method that addresses many of the limitations of MSA-based phylogenetic methods, and computes alignment-free phylogenetic analyses on large sets of proteins that also have long sequences. Unlike other methods, we incorporate the use of standard substitution matrices to improve similarity measures between sequences. Our results show that highly similar phylogenies are produced to existing MSA-based methods with less computational resources required.

ACHIEVEMENTS

  • Short paper and poster: ACB BCB ’13 – ACM International Conference on Bioinformatics, Computational Biology and Biomedicine, Sept 22-25, Washington DC [link] [PDF]
  • Poster: Kalman Research Symposium 2013, April 13, Bucknell University, Lewisburg, PA.

POST GRADUATION UPDATES

Brigitte is pursuing a doctorate at University of Georgia in Bioinformatics, starting Fall 2014

Matthew Segar, ’12

Project: A probabilistic method for assembly of next generation sequencing instrumentation
Duration: Summer 2011 – Spring 2012
Funding: Bucknell PUR, Provost’s Office, CS. Dept Funds

ABSTRACT

With the advent of cheaper and faster DNA sequencing technologies, assembly methods have greatly changed. Instead of outputting reads that are thousands of base pairs long, new sequencers parallelize the task by producing read lengths between 35 and 400 base pairs. Reconstructing an organism’s genome from these millions of reads is a computationally expensive task. Our algorithm solves this problem by organizing and indexing the reads using n-grams, which are short, fixed-length DNA sequences of length n. These n-grams are used to efficiently locate putative read joins, thereby eliminating the need to perform an exhaustive search over all possible read pairs. Our goal is to develop a novel n-gram method for the assembly of genomes from next-generation sequencers. Specifically, a probabilistic, iterative approach will be utilized to determine the most likely reads to join through development of a new metric that models the probability of any two arbitrary reads being joined together. Tests were run using simulated short read data based on randomly created genomes ranging in lengths from 10,000 to 100,000 nucleotides with 16 to 20x coverage. We have been able to successfully re-assemble entire genomes up to 100,000 nucleotides in length.

ACHIEVEMENTS

  • Honor’s Thesis: A probabilistic method for assembly of next generation sequencing instrumentation
    • Matt was awarded the Harold W. Miller prize — a competitive university-wide award given to one or two students at graduation that complete a highly successful honors thesis. CONGRATULATIONS, MATT! The award was well-deserved.
  • Poster (International Conference) – Presented at 20th Annual International Conference on Intelligent Systems for Molecular Biology, ISMB 2012, July 15-17, Long Beach, CA
  • Poster – Susquehanna Valley Undergraduate Research Symposium, August 9, 2011 Geisinger Research, Danville, PA
  • Poster – Sigma Xi Summer Research Symposium, July 27, 2011, Bucknell University, Lewisburg, PA

POST GRADUATION UPDATES

Matt completed a masters in bioinformatics at Indiana University – Purdue University Indianapolis in Spring, 2014. He has now been accepted into the School of Medicine at Indiana University.