Inpactor2 | LTR retrotransposon detector and classificator using Deep

Inpactor2 is a Jupyter Notebook library. Inpactor2 has no bugs, it has no vulnerabilities, it has a Strong Copyleft License and it has low support. You can download it from GitHub.

First, Inpactor2 receives a genome assembly in FASTA format, then each sequence in the input file is split into sections of 50 Kb, without overlapping (stride of 50 Kb). Each section is converted into a 2D-representation using one-hot encoding. Next, a convolutional neural network (CNN) called "Inpactor2_Detect" is used to predict which section contain LTR-RTs and these sections are retained for further analysis. Then, LTR_finder is run to search the beginning and end positions of the previously detected LTR-RT. This step is executed in parallel in order to reduce the execution time. After, a CNN called "Inpactor2_K-mers" is used to count k-mer frequencies in the extracted LTR-retrotransposons. This CNN is intended to extract features required by the next NNs of the pipeline in a time-efficient way. Then, intact sequences are filtered and retained based on the k-mer frequencies and a fully connected neural network (FNN) called "Inpactor2_Filter". Finally, another FNN named "Inpactor2_Class" is used to classify the elements into lineages. Inpactor2 can be executed using several (from one to five) cycles of analysis to detect LTR-RTs divided by the sections generated by the software. Additionally, Inpactor2 can be performed using different structural parameters to filter LTR-RTs, such as minimum and maximum length, LTR beginning with TG and finishing with CA, and target site duplication (TSD) before and after the element.