Systems biology to predict progression and treatment response in M avium complex pulmonary disease

PROJECT SUMMARY Prevalence of pulmonary disease due to non-tuberculous mycobacteria in the United States has doubled in the last 10-years, and Mycobacterium avium pulmonary disease (MAC-PD) is the most common. Treatment requires ≥3 antibiotics for >12 months and efficacy remain low. Given this intense treatment regimen and limited efficacy,

antibiotic therapy is not always advisable. In fact, ~50% of patients will attain culture conversion with observation alone or with bronchial hygiene treatments. There is currently no way to predict which patients will improve with or without antibiotic treatment. The result of this uncertainty is prolonged ineffective treatment which risks side

effects, drive drug resistance and extend treatment duration. We propose an interdisciplinary approach that combines clinical, immunological and pharmacological data. We will integrate these datasets from retrospective and prospective cohorts using machine learning and mechanistic computational simulations with the goal of

developing a disease progression risk score (DP-RS) for patients managed without antibiotics (Aim 1), and a treatment failure risk score (TF-RS) for patients treated with guidelines-based antibiotic regimens (Aim 2). Such risk scores can help to predict patients at risk of disease progression without antibiotics or treatment failure with

antibiotics, so they can be prioritized for closer observation or intensified treatment, respectively. For patients who do not receive antibiotics (Aim 1) we will a) use retrospective clinical data from Indiana, Florida and Minnesota to train and test random forest algorithms to predict disease progression and validate our predic-

tions against prospective clinical cohorts from Florida and Minnesota; b) identify antigen specific T cells re- sponses associated with disease progression; c) predict tissue-level disease progression based on patient-spe- cific T-cell responses using mechanistic computational models; and d) train and test simulation assisted random

forest (SARF) algorithms that integrate clinical, immunological and computational data to predict disease pro- gression and derive the DP-RS. For patients who receive antibiotics (Aim 2) we will a) use retrospective clinical data (Indiana, Florida, Minnesota) to train and test random forest algorithms to predict treatment failure and validate our predictions against pro-

spective clinical cohorts (Florida, Minnesota); b) identify antigen specific T cells responses associated with treat- ment failure; c) identify serum and bronchoalveolar lavage pharmacokinetics associated with treatment failure; d) predict tissue-level treatment response based on patient-specific T-cell responses and pharmacokinetics us-

ing mechanistic computational models; and d) train and test SARF algorithms that integrate clinical, immunolog- ical, pharmacological and computational data to predict treatment failure and derive the TF-RS. Together, this work will provide fundamental insights into clinical, immunological and pharmacological contribu-

tors to MAC-PD disease progression and treatment response, as well as provide predictive risk scores (DP-RS and TF-RS) that will translate into precision medicine management tools, helpful for clinicians and patients.