A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action

Jason H. Yang; Sarah N. Wright; Meagan Hamblin; Douglas McCloskey; Miguel A. Alcantar; Lars Schrübbers; Allison J. Lopatkin; Sangeeta Satish; Amir Nili; Bernhard O. Palsson; Graham C. Walker; James J. Collins

doi:10.1016/j.cell.2019.04.016

A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action

Jason H. Yang
, Sarah N. Wright
, Meagan Hamblin
, Douglas McCloskey
, Miguel A. Alcantar
, Lars Schrübbers
, Allison J. Lopatkin
, Sangeeta Satish
, Amir Nili
, Bernhard O. Palsson
, Graham C. Walker
, James J. Collins

University of Iceland

University of Iceland

Research output: Contribution to journal › Article › peer-review

Abstract

Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated “white-box” biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genome-scale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy. Causal metabolic pathways underlying antibiotic lethality in bacteria are illuminated by a network model-driven machine learning approach, overcoming limitations of existing “black-box” approaches that cannot reveal causal relationships from large biological datasets.

Original language	English
Pages (from-to)	1649-1661.e9
Journal	Cell
Volume	177
Issue number	6
DOIs	https://doi.org/10.1016/j.cell.2019.04.016
Publication status	Published - 30 May 2019

Bibliographical note

Funding Information: The authors thank Ian Andrews, Sarah Bening, and Charley Gruber from MIT; Eric Brown and Madeline Tong from McMaster University; Eytan Ruppin from the National Cancer Institute; Ahmed Badran, Eachan Johnson, and Keren Yizhak from the Broad Institute; Sylvie Manuse from Northeastern University; and Xilin Zhao from New Jersey Medical School for helpful discussions. This work was supported by grant HDTRA1-15-1-0051 from the Defense Threat Reduction Agency (to J.J.C.), grant K99-GM118907 from the NIH (to J.H.Y.), a National Science Foundation graduate research fellowship 1122374 (to M.A.A.), grant U01-AI124316 from the NIH (to B.O.P.), the Novo Nordisk Foundation (to B.O.P.), grants R01-CA021615 and R35-ES028303 from the NIH (to G.C.W.); grant U19-AI111276 from the NIH (to J.J.C.), and support from the Paul G. Allen Frontiers Group , the Broad Institute at MIT and Harvard , and the Wyss Institute for Biologically Inspired Engineering (to J.J.C). G.C.W. is an American Cancer Society professor. Funding Information: The authors thank Ian Andrews, Sarah Bening, and Charley Gruber from MIT; Eric Brown and Madeline Tong from McMaster University; Eytan Ruppin from the National Cancer Institute; Ahmed Badran, Eachan Johnson, and Keren Yizhak from the Broad Institute; Sylvie Manuse from Northeastern University; and Xilin Zhao from New Jersey Medical School for helpful discussions. This work was supported by grant HDTRA1-15-1-0051 from the Defense Threat Reduction Agency (to J.J.C.), grant K99-GM118907 from the NIH (to J.H.Y.), a National Science Foundation graduate research fellowship 1122374 (to M.A.A.), grant U01-AI124316 from the NIH (to B.O.P.), the Novo Nordisk Foundation (to B.O.P.), grants R01-CA021615 and R35-ES028303 from the NIH (to G.C.W.); grant U19-AI111276 from the NIH (to J.J.C.), and support from the Paul G. Allen Frontiers Group, the Broad Institute at MIT and Harvard, and the Wyss Institute for Biologically Inspired Engineering (to J.J.C). G.C.W. is an American Cancer Society professor. Conceptualization, J.H.Y.; Methodology, J.H.Y. S.N.W. M.H. D.M. M.A.A. L.S. and A.J.L.; Investigation, J.H.Y. S.N.W. M.H. D.M. M.A.A. L.S. A.J.L. S.S. and A.N.; Formal Analysis, J.H.Y. S.N.W. and D.M.; Visualization, J.H.Y.; Writing, J.H.Y. G.C.W. and J.J.C.; Resources, J.J.C. B.O.P. and J.H.Y.; Funding Acquisition, J.J.C. G.C.W. B.O.P. and J.H.Y.; Supervision, B.O.P, G.C.W. and J.J.C. J.J.C. is a scientific co-founder and scientific advisory board chair of Enbiotix, an antibiotics startup company. Publisher Copyright: © 2019 Elsevier Inc.

Other keywords

ATP
LC-MS/MS
NADPH:NADP ratio
adenylate energy charge
antibiotics
biochemical screen
machine learning
metabolism
network modeling
purine biosynthesis

Access to Document

10.1016/j.cell.2019.04.016

Cite this

@article{f302fedc69e04e0c8755cf5d3ec1b719,

title = "A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action",

abstract = "Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated “white-box” biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genome-scale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy. Causal metabolic pathways underlying antibiotic lethality in bacteria are illuminated by a network model-driven machine learning approach, overcoming limitations of existing “black-box” approaches that cannot reveal causal relationships from large biological datasets.",

keywords = "ATP, LC-MS/MS, NADPH:NADP ratio, adenylate energy charge, antibiotics, biochemical screen, machine learning, metabolism, network modeling, purine biosynthesis",

author = "Yang, \{Jason H.\} and Wright, \{Sarah N.\} and Meagan Hamblin and Douglas McCloskey and Alcantar, \{Miguel A.\} and Lars Schr{\"u}bbers and Lopatkin, \{Allison J.\} and Sangeeta Satish and Amir Nili and Palsson, \{Bernhard O.\} and Walker, \{Graham C.\} and Collins, \{James J.\}",

note = "Funding Information: The authors thank Ian Andrews, Sarah Bening, and Charley Gruber from MIT; Eric Brown and Madeline Tong from McMaster University; Eytan Ruppin from the National Cancer Institute; Ahmed Badran, Eachan Johnson, and Keren Yizhak from the Broad Institute; Sylvie Manuse from Northeastern University; and Xilin Zhao from New Jersey Medical School for helpful discussions. This work was supported by grant HDTRA1-15-1-0051 from the Defense Threat Reduction Agency (to J.J.C.), grant K99-GM118907 from the NIH (to J.H.Y.), a National Science Foundation graduate research fellowship 1122374 (to M.A.A.), grant U01-AI124316 from the NIH (to B.O.P.), the Novo Nordisk Foundation (to B.O.P.), grants R01-CA021615 and R35-ES028303 from the NIH (to G.C.W.); grant U19-AI111276 from the NIH (to J.J.C.), and support from the Paul G. Allen Frontiers Group , the Broad Institute at MIT and Harvard , and the Wyss Institute for Biologically Inspired Engineering (to J.J.C). G.C.W. is an American Cancer Society professor. Funding Information: The authors thank Ian Andrews, Sarah Bening, and Charley Gruber from MIT; Eric Brown and Madeline Tong from McMaster University; Eytan Ruppin from the National Cancer Institute; Ahmed Badran, Eachan Johnson, and Keren Yizhak from the Broad Institute; Sylvie Manuse from Northeastern University; and Xilin Zhao from New Jersey Medical School for helpful discussions. This work was supported by grant HDTRA1-15-1-0051 from the Defense Threat Reduction Agency (to J.J.C.), grant K99-GM118907 from the NIH (to J.H.Y.), a National Science Foundation graduate research fellowship 1122374 (to M.A.A.), grant U01-AI124316 from the NIH (to B.O.P.), the Novo Nordisk Foundation (to B.O.P.), grants R01-CA021615 and R35-ES028303 from the NIH (to G.C.W.); grant U19-AI111276 from the NIH (to J.J.C.), and support from the Paul G. Allen Frontiers Group, the Broad Institute at MIT and Harvard, and the Wyss Institute for Biologically Inspired Engineering (to J.J.C). G.C.W. is an American Cancer Society professor. Conceptualization, J.H.Y.; Methodology, J.H.Y. S.N.W. M.H. D.M. M.A.A. L.S. and A.J.L.; Investigation, J.H.Y. S.N.W. M.H. D.M. M.A.A. L.S. A.J.L. S.S. and A.N.; Formal Analysis, J.H.Y. S.N.W. and D.M.; Visualization, J.H.Y.; Writing, J.H.Y. G.C.W. and J.J.C.; Resources, J.J.C. B.O.P. and J.H.Y.; Funding Acquisition, J.J.C. G.C.W. B.O.P. and J.H.Y.; Supervision, B.O.P, G.C.W. and J.J.C. J.J.C. is a scientific co-founder and scientific advisory board chair of Enbiotix, an antibiotics startup company. Publisher Copyright: {\textcopyright} 2019 Elsevier Inc.",

year = "2019",

month = may,

day = "30",

doi = "10.1016/j.cell.2019.04.016",

language = "English",

volume = "177",

pages = "1649--1661.e9",

journal = "Cell",

issn = "0092-8674",

publisher = "Elsevier B.V.",

number = "6",

}

TY - JOUR

T1 - A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action

AU - Yang, Jason H.

AU - Wright, Sarah N.

AU - Hamblin, Meagan

AU - McCloskey, Douglas

AU - Alcantar, Miguel A.

AU - Schrübbers, Lars

AU - Lopatkin, Allison J.

AU - Satish, Sangeeta

AU - Nili, Amir

AU - Palsson, Bernhard O.

AU - Walker, Graham C.

AU - Collins, James J.

N1 - Funding Information: The authors thank Ian Andrews, Sarah Bening, and Charley Gruber from MIT; Eric Brown and Madeline Tong from McMaster University; Eytan Ruppin from the National Cancer Institute; Ahmed Badran, Eachan Johnson, and Keren Yizhak from the Broad Institute; Sylvie Manuse from Northeastern University; and Xilin Zhao from New Jersey Medical School for helpful discussions. This work was supported by grant HDTRA1-15-1-0051 from the Defense Threat Reduction Agency (to J.J.C.), grant K99-GM118907 from the NIH (to J.H.Y.), a National Science Foundation graduate research fellowship 1122374 (to M.A.A.), grant U01-AI124316 from the NIH (to B.O.P.), the Novo Nordisk Foundation (to B.O.P.), grants R01-CA021615 and R35-ES028303 from the NIH (to G.C.W.); grant U19-AI111276 from the NIH (to J.J.C.), and support from the Paul G. Allen Frontiers Group , the Broad Institute at MIT and Harvard , and the Wyss Institute for Biologically Inspired Engineering (to J.J.C). G.C.W. is an American Cancer Society professor. Funding Information: The authors thank Ian Andrews, Sarah Bening, and Charley Gruber from MIT; Eric Brown and Madeline Tong from McMaster University; Eytan Ruppin from the National Cancer Institute; Ahmed Badran, Eachan Johnson, and Keren Yizhak from the Broad Institute; Sylvie Manuse from Northeastern University; and Xilin Zhao from New Jersey Medical School for helpful discussions. This work was supported by grant HDTRA1-15-1-0051 from the Defense Threat Reduction Agency (to J.J.C.), grant K99-GM118907 from the NIH (to J.H.Y.), a National Science Foundation graduate research fellowship 1122374 (to M.A.A.), grant U01-AI124316 from the NIH (to B.O.P.), the Novo Nordisk Foundation (to B.O.P.), grants R01-CA021615 and R35-ES028303 from the NIH (to G.C.W.); grant U19-AI111276 from the NIH (to J.J.C.), and support from the Paul G. Allen Frontiers Group, the Broad Institute at MIT and Harvard, and the Wyss Institute for Biologically Inspired Engineering (to J.J.C). G.C.W. is an American Cancer Society professor. Conceptualization, J.H.Y.; Methodology, J.H.Y. S.N.W. M.H. D.M. M.A.A. L.S. and A.J.L.; Investigation, J.H.Y. S.N.W. M.H. D.M. M.A.A. L.S. A.J.L. S.S. and A.N.; Formal Analysis, J.H.Y. S.N.W. and D.M.; Visualization, J.H.Y.; Writing, J.H.Y. G.C.W. and J.J.C.; Resources, J.J.C. B.O.P. and J.H.Y.; Funding Acquisition, J.J.C. G.C.W. B.O.P. and J.H.Y.; Supervision, B.O.P, G.C.W. and J.J.C. J.J.C. is a scientific co-founder and scientific advisory board chair of Enbiotix, an antibiotics startup company. Publisher Copyright: © 2019 Elsevier Inc.

PY - 2019/5/30

Y1 - 2019/5/30

N2 - Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated “white-box” biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genome-scale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy. Causal metabolic pathways underlying antibiotic lethality in bacteria are illuminated by a network model-driven machine learning approach, overcoming limitations of existing “black-box” approaches that cannot reveal causal relationships from large biological datasets.

AB - Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated “white-box” biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genome-scale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy. Causal metabolic pathways underlying antibiotic lethality in bacteria are illuminated by a network model-driven machine learning approach, overcoming limitations of existing “black-box” approaches that cannot reveal causal relationships from large biological datasets.

KW - ATP

KW - LC-MS/MS

KW - NADPH:NADP ratio

KW - adenylate energy charge

KW - antibiotics

KW - biochemical screen

KW - machine learning

KW - metabolism

KW - network modeling

KW - purine biosynthesis

UR - https://www.scopus.com/pages/publications/85065862749

U2 - 10.1016/j.cell.2019.04.016

DO - 10.1016/j.cell.2019.04.016

M3 - Article

C2 - 31080069

SN - 0092-8674

VL - 177

SP - 1649-1661.e9

JO - Cell

JF - Cell

IS - 6

ER -

A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action

Abstract

Bibliographical note

Other keywords

Access to Document

Fingerprint

Cite this