A dipyrimidine sequence library for determining the sequence dependence of cyclobutane pyrimidine dimer formation

The cis-syn CPD is the major photoproduct resulting from UV irradiation that results in C to T mutations found in human skin cancers. In this project, I studied CPD distribution within all possilbe NPyPyN sites. The quantification data given by the project provided some general rules for estimating CPD distribution in duplex DNA.

• Wrote python scripts to generate sequence libraries containing all possible NPyPyN (N is A,T,C,G and PyPy represents CC, TC, CT, TT) sites.
• Sythesized, cloned and PCR amplified one of designed 129-mer sequence which contains all 64 NPyPyN sites.
• Utilized T4-endonuclease and gel electrophoresis to map the CPD distribution within 129-mer under different direct and indirect excitation conditions.
• Analyzed and plotted the final result using python libraries.

K-mer Kernels for Genetic Distance

Phylogenetic tree reconstruction requires construction of a multiple sequence alignment (MSA) from sequences. However, alignment can be quite challenging for very large genomes. An interesting recent approach uses intensive hashing method to count the fraction of all unique k-mers appearing in either genome that are common to both to estimate the evolutionary distance, which avoids the need of assembly and alignment. In this project, I tried to implement this k-mer idea using a more space efficient method based on the 2BWT.

• Wrote python scripts and Java program to process raw data, build 2BWT, count distinct and shared k-mers and generate the genetic distance table.
• Constructed the similar common k-mer table and phylogeny tree as in AAF paper using the simulated reads data provided from paper.
• Applied and tested the method on 4 real genome large data set, and the resulting phylogeny tree similar to the one generated by the program phyloT.

Create Customer Segments

In this project, I successfully employed unsupervised learning algorithms to come up with a decision boundary and build customer segments. The end product provides a basis for segmenting customers and provides a platform to employ A/B tests to enhance profits.

• Reviewed unstructured data to understand the patterns and natural categories that the data fits into.
• Employed PCA to reduce the dimensionality of data and captured 83% of the variability in the dataset.
• Constructed Gaussian Mixture Model and visualized the cluster of customer segments.

Predicting Boston Housing Prices

• Developed a decision tree model to predict the value of a given house in the Boston real estate market using statistical analysis tools .
• Identified the best price that a client can sell their house utilizing machine learning.

Building a Student Intervention System

In this project , I investigated the factors that affect a student's performance in high school.

• Trained and tested several supervised machine learning models(SVM, Decision Tree, Naive Bayes) on a given dataset to predict how likely a student is to pass.
• Selected the best model based on relative accuracy and efficiency.

Train a Smartcab How to Drive

In this project, I applied reinforcement learning to build a simulated vehicle navigation agent to drop off the passenger to the goal state in the shortest time possible.

• Implemented and compared a naive random walk agent and a q-learning driving agent that automatically learned an optimal driving strategy based on rewards and penalties.
• Optimized q-learning parameters for the driving agent to achieve 100% success of reaching destination over 100 trials.

A File Sharing Website

A simple file sharing site that supports uploading, viewing, and deleting files associated with various users.

• On the main login page, you can create a new user and then successfully login using this new username.
• Already existing users have the option to upload, view files as well as delete their username and all files associated with their username.