Grant: DRA : Canada Graduate Research Scholarship - Doctoral Program (CGRS D)
McMaster University
$120,000.00
Active learning-driven discovery of novel antibiotics against Klebsiella pneumoniae
EF 2025-26
Résumé
As bacteria become increasingly resistant to existing antibiotics, once easy to treat infections are becoming difficult to cure. Antimicrobial resistance (AMR) is already associated with nearly five million deaths each year and is projected to cause eight million annually by 2050. This crisis is worsened by the lack of new antibiotics, with no novel class discovered in more than four decades. Among the most dangerous pathogens identified by the WHO is Klebsiella pneumoniae, a frequent driver of hospital-acquired infections that often exhibits multi-drug resistance. Despite the urgent need for new antibiotics, existing discovery methods have delivered few promising leads in recent decades. Artificial intelligence (AI) offers an alternative by predicting antibacterial activity across vast chemical libraries, far beyond what can be tested in the lab. However, current AI models are limited by their training data, which come from antibacterial screens that are costly and yield few active molecules. This restricts the models' ability to learn the features that drive activity. To overcome this, I will use active learning (AL), an AI-guided workflow that addresses this data bottleneck by allowing the model to select its own training data over several rounds. The AL process will begin with a small K. pneumoniae screen to train the model. Based on that training, the model identifies batches of promising compounds from a large chemical library. Those compounds are tested in the lab, and the new results are added back into the training set. Over multiple cycles, this approach improves the model's accuracy and uncovers more hits, all while testing fewer molecules. The top candidates from AL will be characterized in the lab to confirm efficacy and safety, with the goal of identifying novel antibiotic leads. Overall, by coupling AI with lab-in-the-loop feedback, this project provides a faster, more efficient framework for discovering antibiotics against K. pneumoniae and beyond.
Programme
DRA : Canada Graduate Research Scholarship - Doctoral Program (CGRS D)
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
CIHR Funding Decisions Database
Tel que publié
Id
559314
Name
Alexander Jeremie
Region
Ontario
Country
Canada
Orgtype
Institution Paid; Research Institution
Instname2
Infection and Immunity
Theme2
Biomedical
Prcname2
DT1 (Biomedical and/or Clinical Research) CGRS Doctoral
Projecttitle
Active learning-driven discovery of novel antibiotics against Klebsiella pneumoniae
Programtype2
Studentship Programs
Orgnameinp2
McMaster University
Cihramount2
$120,000
Cihrequipment2
$0
Programname2
DRA : Canada Graduate Research Scholarship - Doctoral Program (CGRS D)
Keyworddelim
Active Learning; Antibiotic Resistance; Antibiotics; Artificial Intelligence; Drug Discovery; Machine Learning
Abstract
As bacteria become increasingly resistant to existing antibiotics, once easy to treat infections are becoming difficult to cure. Antimicrobial resistance (AMR) is already associated with nearly five million deaths each year and is projected to cause eight million annually by 2050. This crisis is worsened by the lack of new antibiotics, with no novel class discovered in more than four decades. Among the most dangerous pathogens identified by the WHO is Klebsiella pneumoniae, a frequent driver of hospital-acquired infections that often exhibits multi-drug resistance. Despite the urgent need for new antibiotics, existing discovery methods have delivered few promising leads in recent decades. Artificial intelligence (AI) offers an alternative by predicting antibacterial activity across vast chemical libraries, far beyond what can be tested in the lab. However, current AI models are limited by their training data, which come from antibacterial screens that are costly and yield few active molecules. This restricts the models' ability to learn the features that drive activity. To overcome this, I will use active learning (AL), an AI-guided workflow that addresses this data bottleneck by allowing the model to select its own training data over several rounds. The AL process will begin with a small K. pneumoniae screen to train the model. Based on that training, the model identifies batches of promising compounds from a large chemical library. Those compounds are tested in the lab, and the new results are added back into the training set. Over multiple cycles, this approach improves the model's accuracy and uncovers more hits, all while testing fewer molecules. The top candidates from AL will be characterized in the lab to confirm efficacy and safety, with the goal of identifying novel antibiotic leads. Overall, by coupling AI with lab-in-the-loop feedback, this project provides a faster, more efficient framework for discovering antibiotics against K. pneumoniae and beyond.
Competitiondate
202511
Approvedterm2
3 yrs 0 mth
Pinamesdelim
Alexander, Jeremie
Approvedterm
3 ans 0 mois
Cihrcontribution2
~ $120,000~ $0~