A Computational and Statistical Framework for Screening Novel Antimicrobial Peptides

A COMPUTATIONAL AND STATISTICAL FRAMEWORK FOR SCREENING NOVEL ANTIMICROBIAL PEPTIDES Daniel Paul Veltri, PhD George Mason University, 2015 Dissertation Director: Dr. Amarda Shehu Bacterial resistance to antibiotics continues to be a serious concern worldwide. This has motivated a strong research focus on naturally-occurring antimicrobial peptides (AMPs) as templates for new drug development. To date, experiments in the wet laboratory have characterized thousands of AMPs while generally concentrating on measures of antibacterial activity for natural sequences or peptides designed using a limited number of site-directed mutations. Based on these findings, the computational AMP research community seeks to better understand how biological signals and features relate to antimicrobial activity through the use of machine learning and statistical approaches. In this dissertation, we advance our understanding of the determinants for antimicrobial activity by carefully constructing a set of descriptive features for use in AMP classification models. In addition to using physicochemical features, we also construct new sequence-based features which capture information about distal patterns within a peptide. Comparative analysis with stateof-the-art methods in AMP recognition reveal our methods to be among the top performers while still providing a transparent summary of relative feature importance. Moreover, this dissertation applies our features in a new setting to demonstrate for the first time a computational model to recognize if an AMP may perform better against a representative Gram-positive or Gram-negative bacteria. Work presented is a step forward for in silico research seeking to help guide AMP design in the wet laboratory. Our predictive models are made accessible via AMP Scanner, a new publicly-available web server at: http://www.ampscanner.com. Chapter 1: Introduction 1.1 The Problem of Antibiotic Resistance Drug-resistance in bacteria has been an ongoing problem in hospitals around the world [1–3]. Reports of resistance for the first major clinical antibiotic, penicillin, started to surface in Western hospitals soon after its introduction in the mid 1940’s [4, 5]. More recently, other front-line drugs have started to become ineffective against bacteria such as carbapenem resistant enterobacteriaceae and methicillin-resistant S. aureus (MRSA). A 2013 report from the US Centers for Disease Control and Prevention cites over two million infections and 23,000 deaths a year from antibiotic-resistant bacterial infections in the US alone [6]. In 2014, US President Barack Obama released Executive Order 13676 [7], a “National Strategy for Combating Antibiotic-Resistant Bacteria,” which places a heavy emphasis on promoting new antibiotics and diagnostics. The World Health Organization has also been calling for additional antibiotic development from the medical research community [3]. This dissertation aids in these efforts through the use of computational analysis on naturally-occurring antimicrobial peptides (AMPs) to assist wet laboratory researchers in the discovery of novel treatments effective against drug-resistant bacteria. Given the large financial and regulatory hurdles which have tempered antibiotic development over the past half century, this is also an important academic endeavor [8, 9]. 1.2 Antimicrobial Peptides: Nature’s Solution AMPs are a major component of innate immunity and found across all phyla of life [10]. They comprise a number of protein families which can vary in structure, target specificity