In partial fulfillment of the requirements for the degree of

Doctor of Philosophy in Bioinformatics

in the School of Biological Sciences

Breanna D. Shi

Defends her thesis:

SICKLID: SCIENTIST-INFORMED COMPUTATIONAL KNOWLEDGE LEADING INTERDISCIPLINARY DISCOVERY

Thursday, April 9, 2026

10:00am Eastern

IBB 1128, Suddath Seminar Room

Thesis Advisor:

Dr. Patrick McGrath, Advisor

School of Biological Sciences

Georgia Institute of Technology           

Thesis Co-Advisor:

Dr. Arthur Porto

School of Biology

Florida Museum of Natural History

Committee Members:

Dr. Nicholas Lytle

College of Computing

Georgia Institute of Technology           

Dr. James Stroud

School of Biological Sciences

Georgia Institute of Technology

Dr. Benjamin Freeman

School of Biological Sciences

Georgia Institute of Technology

Abstract:

This thesis transcends the development of isolated computational tools to establish a replicable, socio-technical framework for interdisciplinary science. By progressing from specific technical challenges in cichlid biology to developing expert-informed classifiers and a novel 3D coloration analysis platform, we have engineered scalable processes that bridge the gap between computation and biological inquiry. The creation of the Human-Augmented Analytics Group (HAAG) demonstrates that this framework can effectively mobilize and sustain collaboration, providing a sustainable model for building the expert-centered software essential for tackling complex scientific questions. Ultimately, this work provides not just new methods for analyzing biological data, but a proven blueprint for team science that enhances current research paradigms.

The research begins by addressing a foundational challenge in computational biology: the difficulty of training machine learning models on visually ambiguous species. By developing a track-informed, bimodal classification approach that integrates visual data with motion-based temporal features, this work demonstrates that leveraging the full context of video recordings can dramatically improve classification accuracy. This methodological advance not only solves a specific problem in cichlid sex classification but also establishes a generalizable framework for semi-automated annotation that amplifies limited expert input into large-scale training datasets, reducing the labor bottleneck that has long constrained ecological computer vision.

Building on the principle that computational tools must be shaped by biological expertise, this thesis then addresses a second methodological gap: the lack of standardized protocols for quantifying color on three-dimensional specimens. The resulting platform, developed within an established open-source environment, enables researchers to align texture maps across populations, perform population-level color analysis, and interactively segment anatomical regions of interest. By integrating color analysis directly into a workflow that preserves geometric correspondence, this work transforms color from a trait studied in isolation into a dimension of analysis that can be compared alongside shape, providing a toolkit that puts computational capabilities directly into the hands of the biologists who need them.

The culmination of this thesis is not a single tool but a scalable organizational model for interdisciplinary research. Recognizing that technical solutions alone cannot address the systemic challenges of unsustainable software practices and limited collaborative infrastructure, this work develops and implements frameworks for structuring online research teams, training early-career researchers as effective mentors, and building sustainable operational systems. The evolution of the resulting research group across multiple semesters provides evidence that expert-centered collaboration can be systematically organized to produce high-impact outcomes while simultaneously developing the next generation of interdisciplinary research leaders.