Metagenomics-novel-detection

This package we provide you functions for analysing k-mer frequency profiles with 2 purposes.

1. Comparing raw sequences with embedding sequences derived from existing deep learning based taxonomic classification models.
2. Detecting novel species that are not in the reference database.

API Reference

Distances between species

get_distances_between_species()

A function for calculating distances between unknown species and known species (know what is unknown).

• Input is unknown and known features (k-mer frequency profiles), its labels (species ids), and type of distance used to calculate.
• Output is a DataFrame containing mean of distances between unknown and known species.

Parameter	Type	Description
not_novel	array	The features (k-mer frequency profiles) for known species.
not_novel_species	array	The lebels (species labels) for each row of known species.
novel	array	The features (k-mer frequency profiles) for unknown species.
novel_species	array	The lebels (species labels) for each row of unknown species.
distance	String	The distance metric used to calculate distances between species.

• How to use this function

Takes two dataset of species (features and labels of both known and unknown) and returns the DataFrame containing pair-wise distances between unknown and known species.

d = Distances()
d.get_distances_between_species(not_novel,
                                not_novel_species,
                                novel,
                                novel_species,
                                distance)

An example of output from this function.

Novel Species	Not Novel Species	Mean of Distances
1085644	1270	0.941044
1085644	2021	0.905730
1085644	28104	0.873885
1085644	29581	0.861263
1085644	33028	0.532799
1085644	115561	0.759104
1085644	153493	0.728929
1085644	154288	0.553854
1085644	180957	0.748921
1085644	485870	0.871557
1085644	485898	0.849495

According to this output:

• The species "1270" (known) has the furthest dtistance away from species "1085644" (unknown).
• The species "33028" (known) has the closest distance to species "1805644" (unknown).

Silhouette score

Silhouette score is . . .

get_silhouette_score()

A function for calculating silhouette scores of projected clusters from UMAP.

• Input is projected coordinates and its labels.
• Output is a silhouette score of UMAP's projected clusters.

This function will first feed UMAP's projected coordinates to HDBSCAN to extract clusters labels, then use those labels instead of species labels for calculating silhouette scores.

Parameter	Type	Description
train_embed	array	The projected coordinates of known species form UMAP.
test_embed	array	The projected coordinates of unknown species from UMAP.
n_samples	float	Number of faetures per species.
distance	String	The distance metric for HDBSCAN.

• How to use this function

Takes two dataset of projected coordinates derived from UMAP (known and unknown species) and returns the silhouette score of projected clusters.

s = Scores()
s.get_silhouette_score(train_embed,
                       test_embed,
                       n_samples,
                       distance)

An example of output from this function.
0.8614895