HMM GUIDED MUTATION

Hidden Markov Model Guided Amino Acid Mutagenesis

Running Python Script

Output File System

• Figures related to RMSD indicator data are generated only when HAS_INDICATOR is False.

Overview

1. From MD trajectory data, extract distance measurements between active site AA & substrate atoms.
2. Using the distance data, train Gaussian Emission HMM and label each frame/timepoint of test data with a hidden state.
3. Score each hidden state using indicator data and get emission probabilities (means and standard deviations of features for all states) of the highest scoring hidden state.
4. Suggest AA substitution based on AA properties using the emission probabilities of the high scoring state.

Input: parameters.yml

Output: final_output.csv

1. Current Residue: original residue at the given residue index of FAP
2. Nearby 3 substrate atoms: three substrate atoms closest to the current residue It is in form of python dictionary=> atom index: [atom name, residue_alphacarbon_to_atom_mean_distance, standard_deviation]
3. AA Suggestion and Length: suggested amino acid and the suggested amino acid’s length

Pipeline Description

• bound = 1 standard deviation from mean

Notes

• hmm_guided_mutation script loads MD trajectory data and therefore could take up a lot of RAM.

License

• Project is available under the MIT license found in the LICENSE file.

Expected Model Change Binary Classification

Expected Model Change (EMC) using Logistic Regression as base model

Example config

Dependencies

Comparison of EMC & random sampling metric plots for 2 tasks

1. Stroke prediction using health record data
   
   Link to Source
   
   
2. Anti-bacterial small molecule screening with molecular fingerprint data

Performance comparison to no-sampling LR model

INFANT MICROBIOME HMM

Code designed to run in linux
python 3.7
Ubuntu 22.04.3 LTS


Link to Repository

Background & Purpose

• User can run infant_microbiome_hmm.py to simply run HMM without dividing data into two sets based on given labels.

• User can run infant_microbiome_two_hmm.py if the user has label data that classifies each infant either as GN or GF. This script runs DMM to assign states to all samples in the combined data, then divides the data into GN and GF set to run HMM separately on the two data sets. HMM outputs are saved in separate folders, saving GN HMM output first, then GF HMM output next.

• DMM outputs will be saved in dmm_outputs/dmm_out

• HMM outputs will be saved in hmm_outputs/hmm_out
  • If multiple rounds of code were ran, output of each round will be saved in a different directory with different indexing
  
  
• Sample microbiome abundance data is given at "infant_microbiome_hmm/data/otu_table_sample.csv"

• Sample label is given at "infant_microbiome_hmm/data/class_labels.tsv"

• No test data is provided.

DMM Dependencies

Output File System


Input Data

• Input data is Operational taxonomic unit (OTU) table csv file.

• OTU table in this project contains abundance of bacteria taxa (data type = float)

• The first column of OTU should be the "Taxa" column that contains name of bacterial species (data type = string)

Steps to use the code

Run initiate_hmm.py

• Downloads DMM code zip file
• Extracts DMM zip file
• Saves parameters.txt at current working directory (which shoulod be infant_microbiome_hmm). parameters.txt file must be read either by infant_microbiome_hmm.py or infant_microbiome_two_hmm.py to run DMM and HMM

python3 initiate_hmm.py

• After running initiate_hmm.py, user can type in the parameters that will be used in scripts that run DMM and HMM
  • User must type in path to input data in the parameters.txt file for running the next script
• Default parameter values are set for provided sample data

Run single HMM (a) or two HMM (b)

• The format of the training data and the label file must match the given sample to run these scripts.

(a) infant_microbiome_hmm.py (single HMM output script)

• Preprocesses data 1. Select rows (taxa) with highest variance (User can define fraction of the taxa with highest variance) 2. Select 10 timepoint samples of each infant based on start PMA timepoint 3. Drop infants with more than 5 missing timepoint samples
  • Runs DMM code to assign states to each sample
  • Combines state outputted by DMM code with infant microbiome abundance timeseries train data
  • Runs HMM

• Input:
  
  • parameters.txt
• Output:
  
  • dmm_outputs/dmm_out
  • hmm_outputs/hmm_out
    • HMM output from given dataset

python3 infant_microbiome_hmm.py -p ./parameters.txt

(b) infant_microbiome_two_hmm.py (two HMM output script)

• • Preprocesses data
    1. Select rows (taxa) with highest variance (User can define fraction of the taxa with highest variance)
    2. Select 10 timepoint samples of each infant based on start PMA timepoint
    3. Drop infants with more than 5 missing timepoint samples
  • Runs DMM code to assign states to each sample
  • Combines state outputted by DMM code with infant gut microbiome data
  • Separates data into GN and GF sets based on given label data
  • Runs HMM on GN data set
  • Runs HMM on GF data set

• Input:
  
  • parameters.txt
  • class_labels.tsv
    • must be a tsv file, not csv
• Output:
  
  • dmm_outputs/dmm_out
  • hmm_outputs/hmm_out_1
    • HMM output from GN dataset
  • hmm_outputs/hmm_out_2
    • HMM output from GF dataset

python3 infant_microbiome_two_hmm.py -p ./parameters.txt -l ./data/class_labels.tsv

output_p_of_o.py:

• Preprocesses test OTU_table.csv file same way input data was preprocessed in infant_microbiome_hmm.py / infant_microbiome_two_hmm.py.
  
  • This script does not select rows based on variance. Specific rows (taxa) must be preselected by the user before feeding the test data to this script.
• Calculates P(O) for all the infants in the test data
• Saves ID and correlating P(O) value as a csv file
• Input:
  
  • file path to microbiome abundance timeseries test data
  • directory path to hmm_out
  • path to save output
  • number of states used to run DMM and HMM
  • PMA start date used to run DMM and HMM
• Output:
  
  • csv file with infant id as columns and P(O) (probability of observing data) of each infant as the first row.
    • output_p_of_o.py should be ran using GN and GF HMM outputs
    • Each infant id and P(O) should be compared to make prediction if the infant will be GN or GF

python3 output_p_of_o.py -f data/otu_table_test_data.csv -m ./hmm_outputs/hmm_out_1 -o ./prediction.csv -s 6 -p 196

Notes

• Additional notes are provided in the beginning of the python scripts.
• Caveats to using DMM code:
• If problem occur in running DMM code, Error message will be printed out in the terminal: "Error occured while running DMM"
• This is possibly due to missing library package for compiling DMM C code. If such is suspected, try installing package at the therminal by typing:

• sudo apt-get install libgsl0ldbl

• When the dimension of the input data is too big, error occurs in the DMM code. So, my code is designed to check if the column number goes over 1800. DMM code doesn't run with sample data (shape=(444 x 2997)) included in infant_microbiome_hmm code. It helped to rename the column names to integers, shortening each column name.
• My code initially checks if the column number of the input data goes over 1800. If it does, it trims the columns based on the starting PMA date first before running DMM. If the column number doens't go over 1800, it runs DMM first before trimmming down columns.
• The preset parameters in the parameters.txt file genereated from initiate_hmm.py file are values determined based on the writer's code runs.
• Likelihood of the HMM is calculated using Viterbi path, not the full joint probability.
• Infant label file must be a tsv file not a csv file.

MAZE GAME

Maze Generation

Pygame Framework

PPI Network Motif Finding

Computational Genomics

Network Motif Finding

1. Using predefined motifs: Some known biological motifs are Feed-Forward Loops (FFLs), 3-Cycles, BiFans, Single-Input Module (SIM) and Multiple Input Module (MIM). In particular, we search for FFLs, 3-Cycles and BiFans in a directed version of the graph.
2. Searching for motifs de novo: When interesting motifs are not known, we can use existing motif structures in the network to query the whole network. However, this can be computationally expensive and is generally considered NP-complete. The pattern-growing and pruning method is used here to grow size-k motifs from scratch and query them in the networks.