Ezra Herman



Enrichment Student

Cohort year



Ezra Herman is a PhD student at the University of York, studying the interactions between bacteria and their viruses (known as phages). He obtained his MSc in Statistics at the University of Glasgow and his BSc in Biological Sciences from the University of Edinburgh. His PhD combines laboratory experimentation, bioinformatic analysis and statistical modelling to understand the drivers of bacterial resistance to phages. This is important from a medical perspective, as phages are increasingly being considered as therapeutics to support the shrinking range of effective antibiotics. 

Besides his research, Ezra is interested in data science education. With funding from the Software Sustainability Institute, he wrote a 3-day regression workshop for the Carpentries Incubator. He also regularly instructs with Ed-DaSH, teaching statistics and Snakemake through online workshops. Additionally, Ezra is interested in tool development. As part of his research, he is designing a Snakemake pipeline that will allow microbiologists to explore the immune systems of their bacterial collections.

Research interests

Whenever researchers perform a hypothesis test, they are at risk of obtaining a false result. They may conclude that an effect is present when one does not exist (a Type I error). Alternatively, they may conclude that an effect is absent when one does exist (a Type II error). From a practical point of view, it is important that researchers choose a test that is appropriate to their question, while minimising the chance of a false result. 

Microbiologists commonly measure the ability of a panel of phages to infect a panel of bacteria. This results in a binary matrix, known as a Phage-Bacteria Infection Network (PBIN). Structure in these matrices, such as nestedness, can suggest underlying ecological and evolutionary dynamics. To measure whether nestedness is present, the PBIN is compared to a series of null matrices, derived from the PBIN through a null model. Many null models exist, with varying Type I and Type II error rates. A previous analysis concluded nestedness to be common in PBINs. However, the null model that has commonly been used in PBIN research has a high Type I error rate. Therefore, nestedness may be rarer than is currently thought.

At the Turing, Ezra will be studying the effect of null model choice on nestedness tests in PBINs through a meta-analysis. With this work he hopes to showcase the importance of null model choice and help researchers decide on the most appropriate null model for their PBIN structure tests.