This project classifies breast cancer tumors into two categories, benign or malignant, depending on tumor characteristics. We are using the Wisconsin Diagnostic Breast Cancer dataset obtained from Kaggle.

Attributes include ID number, diagnosis (B = benign, M = malignant), and 30 real-valued input features. This project will focus on 11/32 attributes given in this dataset. Features are computed from a digitized image of breast mass, and the data describes characteristics of the cell nuclei present in an image.

In total, the dataset contains 569 observations and 12 variables. We split the data into two sets to build our models: 70% training data, and 30% test data.

Data Models:

Logistic Regression

K-Nearest Neighbors

Naive Bayes

Neural Nets

Conclusion:

Figure: All models versus real diagnosis values (y)

Our Naive-Bayes classifier had the highest accuracy with 95.91%, followed by KNN at 95.32% accuracy and our neural network model at 94.74%. All three models outperformed the baseline logistic regression model at 92.40%. Despite the high accuracy, some of these errors have a wide variance. Namely, neural nets misclassified several benign tumors as malignant, with high confidence. These false positives would be problematic in a clincial setting. In future work, these models could be adjusted more, tuning parameters with a separate set of cross-validation data to improve test performance.

Skills utilized: R, H2O

Accepted to ACM's SIGCSE 2018 Student Research Competition, Baltimore, Maryland.

Funded by: FURI, Bidstrup Fellowship, ACM

Study Genie is a web-based study tool that aims to explore the role of collaborative learning and note-taking on student performance in lower division computer science courses. Students may write their notes or code snippets in mini notecards, and then assemble them into a cheatsheet for use during an exam.

All student movements are tracked (i.e., notecard clicked, cheatsheet added, copying and pasting vs. typing), allowing my team to analyze and understand how students study, and then compare these statistics with overall student performance in the course and exams.

This application differs from others in that it allows students to share and manipulate their notes on a smaller, topic-based scale, in the form of mini notecards. Deployed on a Windows Server using IIS and HTTP Platform Handler.

Implemented Visualizations:

We produced visualizations split into two categories to represent two types of data: activity and content. Activity-based visualizations (usage breakdown and calendar) reflect a student's actions and behavior while using the application; content-based visualizations (topic frequencies) build models around note content to supplement student learning experience.

Skills utilized: Java, SQL, D3.js, HTML/CSS, JavaScript/jQuery, and more

Goal: identify colon cancer subtypes by clustering gene expression data used by this paper: https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1001453.

Data processing:

mRNA expression profiles of a large series of 443 CC and 19 non-tumoral colorectal mucosas [1]. 585 tissue samples with 54675 probes were measured. Raw data was filtered to keep the 1459 probes with highest variance of expression between samples for clustering

Methods and results:

Gene Cluster 3.0 was used for k-means cluster analysis, with accompanying visualizations generated from Java Tree View. Hierarchical consensus clustering was performed in R. Six subtypes were identified based on gene expression data.

Skills utilized: R, Java Tree View, Gene Cluster 3.0

Voice and tone guides are used to help clients develop better content. These contain short, helpful tips that suggest how to integrate a brand's unique voice into their content online.

I was asked to develop this theme for a digital marketing agency. We intended to build individual guides for each client and then hand them off to help them manage their internet prescence.

Upload new logo through Wordpress theme customizer.

A lot of key site features were editable through Wordpress, so clients would have the flexibility of customizing their voice and tone guide.

Change color scheme through Wordpress theme customizer.

Each section on the content type page is also editable (i.e.: user, tips, user's feelings, and sample copy). These are linked through advanced custom fields.

Skills utilized: HTML/CSS, JavaScript/jQuery, and PHP

200+ attendees, 18 sponsors, 7 partners, $9k+ in prizes.

Desert Hacks was founded at Arizona State University with the goal of building a collaborative and innovative community without boundaries. In February 2017, Desert Hacks welcomed students from around the country to participate in a weekend of learning new concepts, building new projects, meeting new people, and being a part of a new, exciting experience. A majority of our attendees were beginners or students who have never attended a hackathon before.

As a member of the marketing committee, I helped design and develop our website, and ran our social media pages (including Facebook, Snapchat, and Twitter).

A walkthrough of Desert Hacks's home page

Additionally, I designed and developed the live website, which assisted hackers during the event.

Schedule was updated in realtime throughout the event (with JavaScript). Sponsor APIs, prizes, hardware, and FAQs were also included in the live site.

Skills utilized: JavaScript/jQuery, HTML/CSS, PHP, social media

The Fulton Undergraduate Research Symposium at ASU takes place every semester. Hundreds of grant-funded students present their research in the form of a poster to their peers, professors, faculty, and industry professionals. In Spring and Fall 2016, I presented my research on quorum sensing: Examining the natural diversity of quorum sensing for orthogonal pathways

Abstract: Homoserine lactone (HSL) quorum sensing (QS) is a method of bacterial cell-to-cell communication. Bioengineers have incorporated QS into biological circuits, using the bacteria’s signaling pathway as a “bio-wire” to connect steps in logical operations. Complex genetic circuitry is inhibited by pathway overlap, or “crosstalk,” and lack of pathways (only four out of 100s have been used in synthetic systems to date). To fill this gap, we have designed a decoupled system to test pathways for orthogonality (lack of crosstalk) using a sender cell, which carries the HSL synthase, and receiver cell, which carries the QS promoter and regulator. Our research will ultimately expand the synthetic biology toolbox and provide engineers with more flexibility when building complex biological circuitry.

TL;DR: We are looking for more orthogonal, or non-overlapping, quorum sensing communication pathwhays to use as a tool for bioengineers to build complex biological circuitry.

I have been working on this research since March 2015, at The Haynes Lab at ASU. We are working towards publishing our data in the near future.

This app was designed and developed by a team of four in March 2014, for my Advanced Software Development course in high school. The app was submitted to the SPARK App Leage Mobile App Competition, partnered with Gilbert Fire Department and Intel.

The purpose of our app is to inform the citizens of Gilbert about their town's fire amenities and general fire safety tips. Since our audience ranges from children and teenagers to the elderly, our main focus was user-friendless and accessiblity.

We implemented a standard grid layout for the menu with Gilbert social media icons on the bottom.

A main feature of the app that I designed was the Google Maps location integration. This feature utilizes the Google Maps API to locate the fire station nearest the user's location. It also allows the user to see a list of fire stations in Gilbert, and their distance from the user's location.

The entire app was built using MIT App Inventor, a GUI that utilizes text-based building blocks to create mobile apps.

A modified version of the mobile application, inspired by our original app, is available in the Google Play Store!

Awards: "Best Overall App", "Girls in STEM"