2026 Cheriton Research Symposium

Lila Kari, Cheriton Faculty Fellow | DC 1302

Beyond Ancestry: Learning Environmental Signatures from DNA

Genomes are usually read as records of ancestry and evolution. Could they also carry other information — for example, an imprint of the environments in which organisms live? If so, extremophiles — organisms thriving under extreme conditions — offer the best chance of detecting it.

We develop and adapt computational methods to detect environmental signals in genomic data, and identify the short DNA patterns underlying them using supervised machine learning, clustering, and structural topic modelling (a natural language processing method). Across nearly 600 bacterial and archaeal genomes, we uncover striking signatures of temperature adaptation, that persist even across their deep evolutionary divide: heat-adapted organisms favor C/G-rich and repeating CG patterns, while cold-adapted organisms favor A/T runs.

Perhaps most surprisingly, the environmental signature is pervasive across the genome rather than confined to particular genes. The associated DNA motifs point to sequence-dependent physical properties of DNA — stability, stacking, and flexibility — suggesting a physical basis for this pervasiveness. Computational analysis of short DNA “words” thus reveals an environmental component of genome organization, complementing the familiar signal of ancestry and offering new insight into life at its limits.

Florian Kerschbaum, Cheriton Faculty Fellow | DC 1302

Privacy-Preserving Collaborative Analytics

Artificial intelligence enables us to extract valuable insights from data, but that data is often siloed across different organizations or entities. This creates a difficult trade-off: either keep the data private and forgo potentially valuable analyses or share it and risk revealing sensitive information. Fortunately, there is another option: performing data analysis directly on encrypted data. While naïve approaches can be prohibitively expensive in terms of computation and communication, carefully designed protocols can significantly reduce these costs.

In this talk, I will present several use cases for privacy-preserving data analysis and, using privacy-preserving k-means clustering as a concrete example, demonstrate how clever protocol design can make secure analyses of large datasets substantially more practical.

Khuzaima Daudjee, Cheriton Faculty Fellow  | DC 1302

Toward Dynamic Data Systems

We have long asked two related questions: given a workload, how should system resources be managed? And given system resources, how should we handle workloads? I will present research toward answering these two questions, drawing on examples from our systems work to illustrate how resource management and workload execution can adapt to changing demands and constraints.

Lunch | DC 1301 (fishbowl)

By invitation