incorporate hunipie predictive model

[ArtPoon]

• we have previously run a one-off analysis of cluster data using @ebrintn 's hunipie model (operates on bootstrap consensus Newick trees generated for each lineage)
• the above method needs to be incorporated into dev branch or some new branch
• the predicted number of infections (I) will need to be passed to the front-end in the same data structure that we are currently passing lineage-level stats, such as number of genomes (?) (dbstats.json)
• need to add an option to drop-down to colour tree by predicted number of infections (make this default?)

[ArtPoon]

@SandeepThokala to handle integration of existing code into backend @bonnielu to implement front-end (colouring lineage tree)

[ArtPoon]

• trees are being reconstructed for random samples of the data (feature vectors) in https://github.com/PoonLab/covizu/blob/e444b881182bd188c1a22b2a7d01f00907038709/covizu/clustering.py#L65
• this returns a path to a file containing the replicate bootstrap trees that have been serialized into Newick tree strings
• this clustering.py script is called in batch_utils.py: https://github.com/PoonLab/covizu/blob/e444b881182bd188c1a22b2a7d01f00907038709/covizu/utils/batch_utils.py#L137
• the trees are read from the file here: https://github.com/PoonLab/covizu/blob/e444b881182bd188c1a22b2a7d01f00907038709/covizu/utils/batch_utils.py#L223
• this is the ideal location to incorporate Erin's method: https://github.com/PoonLab/covizu/blob/e444b881182bd188c1a22b2a7d01f00907038709/covizu/utils/batch_utils.py#L241-L246 because it has access to the bootstrap trees
• we could pass the results here: https://github.com/PoonLab/covizu/blob/e444b881182bd188c1a22b2a7d01f00907038709/batch.py#L157
• and then we could extract the results from the result object and deposit them into db_stats here: https://github.com/PoonLab/covizu/blob/e444b881182bd188c1a22b2a7d01f00907038709/batch.py#L169-L193

[ArtPoon]

For now, let's use the original hunepi model that was trained on the consensus trees. Later we will probably want to use a modified model trained on the individual bootstrap trees.

[ebrintn]

I added the predictive models to the HUNePi git repository in a folder called general_linear_prediction_models. All reads contain lists of models using different summary statistics. Use the summary statistics in the list labelled HUNePi

There are four sets of .rds files in the folder:

• infections_increasing_model_comparisons : models for whether infections are increasing or decreasing
• max_infections_model_comparisons: models of maximum infections when decreasing - these models suck, don't use
• num_infections_decreasing_model_comparisons: models of infections when decreasing - these models also suck, don't use
• num_infections_model_comparisons: models of infections when increasing - these are good to use :)

[ArtPoon]

@GopiGugan noted that we are missing the calculation of Pi statistics, pending meeting with @ebrintn for clarification