This is a project that uses routinely collected clinical data to predict mortality in patients with serious mental illnesses. We have developed a human-interpretable machine learning technique to make predictions for mortality.

Files and Layout

* code folder
    
  * has code 

* manuscript folder

    * has manuscript and supplementary material

* installation

    * R --no-save < INSTALL_MANY_MODULES.R
    * python -m pip install -r requirements.txt
    
* data extraction from clinical database

    * data_saver.rmd
    
    * usage:
    
        * run or knit in R Studio
    
* classical statistical models

    * analysis_mortality_cpft_metafeatures.rmd
    
    * usage:
    
        * run or knit in R Studio
        
    * advanced_analysis_mortality_cpft_metafeatures.rmd
    
            * advanced code for follow-up paper
    
* machine learning models

    * mixed_datatype_autoencoder_keras_MLP_multicategorical4_column_deleted_contrast.py
    
    * usage:
    
        * python mixed_datatype_autoencoder_keras_MLP_multicategorical4_column_deleted_contrast.py df_med_suicide_features_lengthened_final_ml_f20_FULL.csv  drug_suidata_autoencoder_f20_MLP_FULL  0.5
        
        * python mixed_datatype_autoencoder_keras_MLP_multicategorical4_column_deleted_contrast_bootstrap.py df_med_suicide_features_lengthened_final_ml_f20_FULL.csv  drug_suidata_autoencoder_f20_MLP_FULL  0.5

        * python advanced_mixed_datatype_autoencoder_keras_MLP_multicategorical4_column_deleted_contrast.py df_med_suicide_features_lengthened_final_ml_f20_FULL_column_deleted.csv  drug_suidata_autoencoder_f20_MLP_FULL_column_deleted_advanced  0.5
            
        * advanced_analysis_mortality_cpft_metafeatures.rmd
    
            * advanced code for follow-up paper
            
        * python ImmuneModel_nomovement_AIS_COMPLETELY_GENERIC_healthcare_bioAI.py tcell_survival_ImmuneModel_nomovement_AIS_COMPLETELY_GENERIC_healthcare_bioAI  tcell_matches_ImmuneModel_nomovement_AIS_COMPLETELY_GENERIC_healthcare_bioAI 1000 False False 100 False   50 1000 2 nocheckauto turnstile notcrseq
            
            * advanced code for follow-up paper data augmentation
            
        * parse_for_negative_selection.R
        
            * advanced code for follow-up paper data augmentation    
            
* synthetic_data

    * code for synthetic data
    
            * autoencoder and feedforward neural network on synthetic data
    
            * calculating and testing standardised mortality ratios
    
* public_data

    * has public data on mortality rates
    
* other functions and helper scripts (also in functions folder)

    * convert_aupr_to_ggplot.R
        * makes AUPR plots in ggplot
    * datetimefunc.R
        * miscellaneous datetime functions from rlib 
    * feature_to_metafeature_mapping.tsv
        * maps features to metafeatures
    * short_descriptions_icd10.tsv
        * stores short text descriptions for some ICD-10 codes
    * chebi_ontology.csv
        * stores drug categories and groupings
    * cam_project.bib
        * bibliography file

Installations

Install Anaconda (Python distribution)

and install all packages by running the following command at the Terminal

pip install -r requirements.txt

For the R packages, install R and run the following command

R --no-save < INSTALL_MANY_MODULES.R

Citation and manuscript

 Generating Complex Explanations for Artificial Intelligence Models: An Application to Clinical Data on Severe Mental Illness, Soumya Banerjee, Life, 14(7), 807, 2024

 https://www.mdpi.com/2075-1729/14/7/807

Contact

* Soumya Banerjee
* sb2333@cam.ac.uk