top of page

Linking genotype to phenotype

deltaBS

Identifying functionally significant genetic variation

 

I work to develop new methods for identifying genetic variation that impacts phenotype. deltaBS allows the pooling of all mutations in a protein coding gene, and weighting of these mutations by their predicted functional impact. 

​

This approach simplifies tests for the association of genetic variation and phenotypes, and allows the incorporation of rare variants into analyses. It also allows predictive methods to process new mutations which haven't been seen in training data in a sensible way. 

deltaBS_custom.png
Learning

Understanding what's "learnable"

An open question in genomics is how many samples we need to collect and sequence to really understand a trait. 

​

For a few key pathogens, I'm collating large collections of genomes, comparing what we already know about resistance to what's present in these collections, and using this to assess how much we could have learned using naive statistical methods and the collections we have today. 

​

This can help us understand when our ability to learn is still limited by our number of samples, and when collecting more samples won't help, and you need to try a new approach. 

candidate_hits.png
LCRFAZM.png
LCSVMAZM.png
Prediction

Predicting phenotype

I'm developing machine learning algorithms to predict phenotype from whole-genome sequencing data. 

​

In addition to developing algorithms for accuracy, I'm also looking at better ways of coding genetic information, to gain more insight from the data we have.

​

I'm also working on more responsible forms of cross-validation, which avoid over-optimistic reporting of accuracy. 

​

The findings below were presented at the 2019 European Congress for Clinical Microbiology and Infectious Disease, indicating that our current methods for cross-validation don't adequately account for relatedness between samples, resulting in over-optimistic reporting of samples. 

AZM_train_test.png
ML_CV_SaArtboard 1_2x.png
bottom of page