Computational Biology

Mitochondrial Segmentation & Damage Analysis

Analysis of individual images collected in microscopy experiments, is tedious and ultimately subjective. In order to successfully analyse the results from a series of labs we performed involving florescent microscopy to measure the toxicity of certain substances to cells, we designed a sufficiently accurate computer vision algorithm to segment mitochondrial images, and quantify damage done to the cells. Due to the relative uniformity in our images, our segmentation approach was able to draw from many well researched ideas in computer vision such as canny edge detection, morphology and difference of gaussians to quickly and accurately approximate the Laplacian of the input image.

Machine Learning Algorithm for Primer Melting Point Prediction

Polymerase chain reaction (PCR) is a widely used technique in biology to selectively amplify certain types of DNA to make identification simpler. A core concept to PCR, is that of thermal cycling — a technique used to induce a sequence of key biological reactions, such as the activation of certain enzymes critical to DNA replication via cycling between heating and cooling. Specifically, the first step of PCR involves separating the two strands of the target DNA by raising the temperature to induce DNA melting. To successfully plan such experiments, understanding the melting point of primers is critical. We wrote a machine learning algorithm which takes a {C,G,T,A}-sequence and outputs the predicted melting point by generating a feature vector of measurements like GC Content, GC Clamp, Number of {A, T, C, G}2-repeats, etc. and employing this in a regression task.

Investigating Fish Mislabeling in Pittsburgh Using PCR

Previous studies have shown that fish mislabeling is a frequently occurring practice across the United states. Fish mislabeling has numerous, negative implications towards both consumers and businesses – deceiving both parties into overpaying for lower quality fish. This lower quality substituted fish may contain unexpected nutritional properties and can cause of health problems for at risk consumers. In this investigation we employ PCR to perform an exploratory study to quantify the degree of fish mislabeling within the local Pittsburgh area. We purchased tuna, yellow tail and salmon samples from three diverse sushi distributors (Entropy, Sushi Fuku, and Mount Everest Sushi) in the Carnegie Mellon University area. The samples were sterilized and then underwent DNA extraction and PCR.

Modeling Early-Stage Microscopic, Avascular Tumor Growth with Hybrid Probabilistic Cellular Automata

In my term project for symbolic programming, I studied computational cancer modelling, and compared the results of various simple cancer models heavily employing topics in pure mathematics. Throughout the development of the project, I surveyed a variety of cancer modelling techniques, and designed modern extensions for some of the simpler models I encountered. One of my favourite results, originally studied in A Cellular Automaton Model of Cancerous Growth, was how well a very simple model based on probabilistic cellular automata could mimic models based on differential equations such as the Gompertz model. Ultimately, I reconfirmed the findings that these models align well, employing much more advanced hardware, and techniques to run simulations at a much larger scale than the original paper was able to do.