Antibiotic Resistance Determination Utilizing Machine Learning

摘要 我们的长期目标是通过开发准确和可解释的预测来解决抗生素耐药性问题 机器学习模型，可以用于临床，以加快细菌感染患者的护理。 目前诊断细菌感染的方法依赖于首先从收集的样本中培养病原体。 然后进行各种表型测试，以确定特定细菌分离物的抗生素 敏感的或抗拒的。在许多情况下，这一过程可能需要几天时间才能完成。开发一种准确的方法来 利用全基因组测序数据预测抗生素耐药性，而不需要表型测试 这个项目的总体目标。我们的团队应用了深度学习和云计算的最新进展。 我们将追求以下具体目标：1)管理大型数据集并开发深度学习预测 对多种细菌种类和抗生素组合具有最先进精度的模型；2) 为慢性感染开发个性化的机器学习模型；3)创建开源的可扩展用户- 为广大研究社区提供友好的资源。这项工作的成功完成将为 我们诊断细菌感染的方式发生了范式转变，并加快了提供正确的 针对特定病原体的抗生素。

Summary Our long-term objective is to tackle antibiotic resistance by developing accurate and interpretable prediction machine learning models that could be used clinically to speed up the care of patients with bacterial infections. Current approaches to diagnosing bacterial infections rely on first culturing a pathogen from a collected specimen followed by a variety of phenotypic tests to determine what antibiotic a particular bacterial isolate would be sensitive or resistant to. This process can, in many cases, take days to finish. Developing an accurate way to predict antibiotic resistance utilizing whole-genome sequencing data without the need for phenotypic testing is the overall goal of this project. Our team applies the latest advances in deep learning and cloud computation. We will pursue the following Specific Aims: 1) Curate a large dataset and develop a deep-learning prediction model with state-of-the-art accuracies for a wide range of bacterial species and antibiotic combinations; 2) Develop personalized machine learning models for chronic infections; 3) create open-sourced scalable user- friendly resources for the broad research community. The successful completion of this work will provide a paradigm shift in the way we diagnose bacterial infections and speed up the time to providing the correct antibiotic for a specific pathogen.