SARS-COV-2 genome analysis using Big Data algorithms in order to find clusters of similar mutations that belongs to different clades which mutate together and generate the correspondent clade.
-
Building a mutation profile (matrix rows: patients, columns: mutations).
-
Generation of a network of similarities between mutations for each clade present in at least 1000 records.
-
Generation of networks (for each clade) representing sets of frequently similar mutations (a triple of mutations which are always connected to others four).
-
For each pair of clades, I calculate the similarity between clusters of frequently similar mutations, in order to study the tendency with which these clusters appear from variant to variant.
(the dataset is private) Each dataset file contains a matrix where each row corresponds to a patient affected by COVID-19 and each column contains various parameters relating to the sample collected, like:
- Name (String): sample name
- Accession (String): missing info
- Geo_Location (String): geographical position where the sample was collected.
- Country (String): country where the sample was collected.
- Collection_Date (Date): sample collection date.
- Host (String): species of the host on which the sample was collected (man...etc)
- Isolation_Source (String): pickup method.
- clade (String): sample virus covid variant (Omicron, 20A, etc...)
- [E.__76...- S.Y904H] (Int): series of binary columns with binary values 0/1 indicating the absence/presence of the particular mutation named by the name of the column.
- Step one: Data Acquisition and Description (Dask and Pandas libraries).
- Step two: Grouping records by the "clade" attribute (covid variant).
- Step three: Filtering the clades which are present in a least 1000 records.
- Step four: Dictionary Creation {Clade: Patient-mutations}.
- Step five: Min-Hashing, Locality Sensitive Hashing (LSH) and Jaccard similarity in order to detect similar covid mutations.
- Step six: Creation of the multilayer network. For each covid clade I build a graph where the nodes are the mutations present in that particular variant and an edge is present between two mutations if their Jaccard similarity is greater than or equal to a threshold equal to 0.3
- Step seven: Arena3D graphs visualization.
- Step eight: A-Priori Association Rules algorithm in order to search for clusters of three mutations frequently associated with others four.
- Step nine: Find the similar clusters of mutations using a Jaccard threshold of 0.75.
The resulting dataframe has the following format:
[Clade 1, Clade 2, C1_itemset, C2_itemset, jaccard_index]
where:
- Clade1 and Clade 2 are the clades to which the similar clusters belong.
- C1_itemset and C2_itemset are the two clusters of seven mutations that pass the threshold.
- jaccard_index is the value of the Jaccard similarity calculated on the two clusters.
