Kumail Alawa

Bioinformatics Specialization · Cohort 2024

Evaluating Long-Read Breakend Calls

Structural Variants (SVs) are variations (like insertions, deletions, inversions, duplications) of length ≥50 base pairs that are found when comparing a sample genome to a reference genome. SVs contribute significantly to human genetic diversity and disease, and they are one of the main goals of long-read sequencing projects. The ongoing All of Us Long Reads project is sequencing ~1000 PacBio CCS samples at 8x coverage. Each sample is analyzed with multiple state-of-the-art SV callers, which, in addition to standard SV calls, emit also a few breakend (BND) calls per sample. BND calls are often discarded because they are believed to be unreliable, but they are potentially useful for representing complex SVs. This work aims at building a machine learning classifier that distinguishes between true and false BNDs by leveraging BND-specific features. We also focus on interpreting the complex variants represented by some true BND calls. We show that true BND calls can be successfully separated from false BND calls (AUC = 0.9). Moreover, we show that, although variant callers are incapable of fully capturing complex variants with BNDs, some of the BND calls they emit are informative.

Read the final report (PDF)

See it in the program report (PDF)

Artificial Intelligence Specialization · Cohort 2023

Utilizing Deep Learning to Produce Audible Arabic Summaries

Lakhes is an AI-based summarization and text-to-speech project that is designed to automate the process of summarizing Arabic text and converting summaries into audible content. By leveraging natural language processing techniques and text-to-speech technology, with the added feature allowing the user to listen to the summary in their favorite voice, the project aims to enhance accessibility and convenience for content consumers in the Arab world.

See it in the program report (PDF)