Design, Screening, and Testing of Non-Rational Peptide Libraries with Antimicrobial Activity: In Silico and Experimental Approaches

Paola Ruiz Puentes, María C. Henao, Carlos E. Torres, Saúl C. Gómez, Laura A. Gómez, Juan C. Burgos, Pablo Arbeláez, Johann F. Osma, Carolina Muñoz-Camargo, Luis H. Reyes, Juan C. CruzAntibiotics 2020 · OpenReview

TL;DR

The rise of antibiotic resistance demands new therapeutic strategies. This paper proposes a novel, four-stage workflow that synergistically combines artificial intelligence (deep learning), molecular dynamics simulations, microbial surface display, and microfluidics to accelerate the discovery and screening of antimicrobial peptides (AMPs). This integrated approach aims to overcome the limitations of traditional methods and efficiently identify promising AMP candidates.

What Problem Does This Paper Address?

The escalating threat of antimicrobial resistance (AMR) is a major global health crisis, rendering existing antibiotics ineffective and leading to increased mortality and healthcare costs. The pharmaceutical industry's dwindling pipeline of new antibiotics exacerbates this problem. Antimicrobial peptides (AMPs), naturally occurring defense molecules, offer a promising alternative due to their diverse mechanisms of action, which may reduce the likelihood of resistance development. However, discovering effective AMPs is a significant challenge, often involving time-consuming and resource-intensive screening of vast peptide libraries. This paper addresses the need for a more efficient, reliable, and cost-effective method to identify and validate AMPs, particularly those that can combat resistant pathogens.

Methods

The authors propose a four-stage workflow that integrates cutting-edge technologies:

  • In Silico Strategies:
    • Deep Learning (DL): Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs), are employed to analyze large databases of peptide sequences. These models learn complex patterns and features within sequences to predict potential antimicrobial activity. This approach aims to overcome limitations in traditional methods that struggle with multi-dimensional data analysis in large datasets.
  • Molecular Dynamics (MD) Simulations: Candidate AMPs identified by DL are then subjected to MD simulations. This technique models the dynamic interactions between peptides and lipid bilayers (mimicking cell membranes) at an atomic level. The goal is to elucidate the mechanism of action, predict membrane disruption capabilities, and assess potential toxicity.
  • Experimental Strategies:
    • Microbial Surface Display: Promising AMP candidates are recombinantly expressed on the surface of microorganisms (bacteria or yeast). This allows for the display of a large number of peptides in a functional context.
  • Microfluidics: High-throughput screening is performed using microfluidic platforms, specifically droplet-based systems. These systems enable the rapid testing of millions of individual peptide-displaying microorganisms against target microbes, allowing for efficient identification of AMPs with desired lytic abilities.

The paper outlines both a "Rational Design" workflow (DL -> MD -> Surface Display -> Microfluidics) and a "Non-Rational Design" workflow (Surface Display -> Microfluidics -> DNA Sequencing), emphasizing the integration of these technologies.

Datasets

The paper discusses the importance of databases for deep learning training, mentioning examples like the Antimicrobial Peptide Database (APD), Collection of Anti-Microbial Peptides (CAMPS), Anuran defense peptides (DADP), and the CAMEL database. However, it does not explicitly detail specific datasets used for training their proposed models or benchmarks for evaluation.

Key Insights and Ideas

  • Synergistic Integration: The core idea is the powerful synergy achieved by combining AI-driven prediction, physics-based simulation, and high-throughput experimental screening. Each technology addresses limitations of the others, creating a more robust discovery pipeline.
  • Beyond Sequence Similarity: Deep learning models can identify functional features in peptide sequences that go beyond simple sequence similarity, potentially uncovering novel AMP scaffolds.
  • Mechanism-Driven Design: MD simulations provide crucial mechanistic insights into how peptides interact with membranes, enabling a more informed selection of candidates and potentially predicting toxicity.
  • High-Throughput Experimental Validation: Microfluidics offers an unprecedented scale for experimental validation, drastically reducing the time and cost associated with screening large peptide libraries.
  • Addressing the "Black Box" of AMPs: By combining computational and experimental approaches, the authors aim to move beyond simply identifying active peptides to understanding why they are active.

Key Results

The paper is primarily a review and proposal of a framework. It does not present specific experimental results or quantitative performance metrics for the proposed workflow. However, it references numerous studies that have successfully employed individual components of the proposed methodology, demonstrating their potential:

  • Deep learning models have shown success in predicting various peptide properties and activities.
  • Molecular dynamics simulations have provided detailed insights into peptide-membrane interactions and mechanisms of action.
  • Microfluidic platforms have demonstrated high-throughput screening capabilities for various biological assays, including those relevant to antimicrobial activity.
  • Table 1 lists several peptides that have been designed and evaluated using MD and DL, with reported antimicrobial activity.

Limitations and Future Work

  • Computational Challenges: The authors acknowledge that state-of-the-art in silico algorithms face challenges in handling the vastness of multi-space data distributions in extremely large databases.
  • Experimental Validation Bottlenecks: While the proposed framework aims to accelerate discovery, the experimental phases (surface display and microfluidics) still require careful optimization and can be resource-intensive, though significantly less so than traditional methods.
  • Generalizability: The paper suggests that models trained for one specific bioactive peptide property could be transferred to other bioactivities by retraining deeper layers, indicating a future direction for developing more versatile predictive models.
  • Broader Applications: The authors believe their proposed approach opens opportunities for searching and screening bioactive peptides for applications beyond antimicrobial activity, such as in neurodegenerative disorders or autoimmune conditions.
  • Further Refinement: The paper highlights the need for robust and stringent metrics to ensure the reliability of computational predictions and avoid overoptimistic results, especially with imbalanced datasets.

Summary updated

Abstract

One of the challenges of modern biotechnology is to find new routes to mitigate the resistance to conventional antibiotics. Antimicrobial peptides (AMPs) are an alternative type of biomolecules, naturally present in a wide variety of organisms, with the capacity to overcome the current microorganism resistance threat. Here, we reviewed our recent efforts to develop a new library of non-rationally produced AMPs that relies on bacterial genome inherent diversity and compared it with rationally designed libraries. Our approach is based on a four-stage workflow process that incorporates the interplay of recent developments in four major emerging technologies: artificial intelligence, molecular dynamics, surface-display in microorganisms, and microfluidics. Implementing this framework is challenging because to obtain reliable results, the in silico algorithms to search for candidate AMPs need to overcome issues of the state-of-the-art approaches that limit the possibilities for multi-space data distribution analyses in extremely large databases. We expect to tackle this challenge by using a recently developed classification algorithm based on deep learning models that rely on convolutional layers and gated recurrent units. This will be complemented by carefully tailored molecular dynamics simulations to elucidate specific interactions with lipid bilayers. Candidate AMPs will be recombinantly-expressed on the surface of microorganisms for further screening via different droplet-based microfluidic-based strategies to identify AMPs with the desired lytic abilities. We believe that the proposed approach opens opportunities for searching and screening bioactive peptides for other applications.
BibTeX
@inproceedings{puentes2020design,
  author = {Paola Ruiz Puentes and María C. Henao and Carlos E. Torres and Saúl C. Gómez and Laura A. Gómez and Juan C. Burgos and Pablo Arbeláez and Johann F. Osma and Carolina Muñoz-Camargo and Luis H. Reyes and Juan C. Cruz},
  title = {{Design, Screening, and Testing of Non-Rational Peptide Libraries with Antimicrobial Activity: In Silico and Experimental Approaches}},
  booktitle = {Antibiotics},
  year = {2020},
  url = {https://openreview.net/forum?id=wkQUjnxUNb},
}
Figure