Skip to content
- Machine Learning for Remote Sensing of Unresolved Resident Space Objects — In collaboration with Dr. Miguel Velez-Reyes in the Electrical and Computer Engineering Department, we are using deep neural networks to process ground-based hyper spectral images of space objects.
- Using Neural Networks to Encode Minimizer Schemes — Since we know that the best performing minimizer ordering tend to be difficult to store and use in practice, we want to use neural architectures to encode known good orderings in compact space with low lookup times.
- Multiple Sequence Alignment Accuracy Prediction — We are extending the Facet (Feature-based Accuracy Estimator) model to use both a linear regression optimizer to find functional coefficients and neural networks to better predict multiple sequence alignment accuracy. These new training methods allow for the exploitation of much larger datasets, and in turn provides higher accuracy parameter advising choices.
- NGS Read Alignment Advising — Using our previous experience with parameter advising on existing bioinformatics tools, we are currently working with a set of NGS aligners to make input-specific parameter choices for this problem. Advances on this topic would improve downstream analysis on many fronts, including out previously studied work on transcript assembly.
- Structure-Aware Protein Multiple Sequence Alignment Advising — The beta version of the Opal alignment tool (opal.cs.arizona.edu, GitHub: github.com/deblasiolab/Opalv3) allows for the inclusion of predicted protein secondary structure when constructing multiple sequence alignments. This extension was not included in our initial study of the impact of parameter advising on alignment.
Information on past projects can be found at dandeblasio.com