PISAD: De novo peptide design for target protein with ...

Abstract

Rapid identification of peptides that bind specifically to a target protein is essential for disease diagnostics and drug development. However, de novo design of peptides without any prior structural knowledge remains a long-standing challenge. Herein, we present Peptide Iterative design with Stochastic Algorithm and Docking (PISAD), a de novo peptide design method, which combines an iterative stochastic searching algorithm with docking assessment. The searching algorithm simulates the evolution of peptide sequences by introducing mutations and crossovers iteratively. After every round of evolution, the peptide sequences undergo docking assessments with the target protein based on the structural prediction of AlphaFold2. We demonstrated PISAD's efficacy by designing peptides targeting four proteins, namely ARF6, ARF1, TGF-β1, and IL-6. For each target, PISAD managed to output peptides with ideal binding affinity within only four iterations of evolutions, and no more than 1250 sequences were assessed. Particularly, the best-performing peptide achieved a KD value of 3.4 nM with ARF6, which has been further experimentally validated. These results demonstrate the efficiency and accuracy of PISAD, which may serve as a universal tool for rapid de novo design of peptides targeting specific proteins.

Introduction

Peptides are widely used as theragnostic agents due to their small size, high selectivity, complementary structures, and low toxicity (Pandey et al., 2021). However, traditional peptide design and screening methods are often time-consuming, labor-intensive, and expensive (Bozovicar and Bratkovic, 2019). In recent years, advancements in computational strategies, such as genetic algorithms (Porto et al., 2018b), pattern insertion (Porto et al., 2018a; Tsai et al., 2022), and deep-learning-based architectures (Das et al., 2021; Lei et al., 2021; Wang et al., 2021a), have been widely used to assist sequence-based design.

Despite the advancements, existing methods still face several limitations. Most methods rely heavily on existing databases or predefined templates to generate peptide sequences. While these methods can leverage existing knowledge and data, they severely restrict the exploration of entirely new peptide sequences, making it difficult to discover novel, non-natural peptide motifs (Hashemi et al., 2021; Nagarajan et al., 2019). Meanwhile, non-deep learning-based methods often lack structural complementarity information when predicting binding affinity, leading to inaccurate and noisy results. Structural complementarity is a critical factor in molecular interactions, and the absence of this information can significantly affect the accuracy and specificity of peptide-target protein binding (Huang, 2014; Vanhee et al., 2011). Moreover, the optimization processes are complex and time-consuming, requiring multiple iterations and extensive experimental validation. This not only increases the time and cost of research but also limits the practicality of these methods (Bozovicar and Bratkovic, 2019; Torres et al., 2018). In addition to these challenges, while deep learning-based methods have shown great promise in predicting the physicochemical properties of peptides and designing functional proteins (Das et al., 2021; Wang et al., 2021a), they often require extensive computational resources, which can be prohibitive for large-scale applications. For example, RFDiffusion (Watson et al., 2023), although effective in designing functional proteins, demands significant computational power and is not well-suited for large-scale peptide design tasks.

To address these limitations, we propose an iterative feedback optimization method called Peptide Iterative design with Stochastic Algorithm and Docking (PISAD), which enables rapid and large-scale de novo peptide design without fundamental structural knowledge. PISAD combines differential evolution (DE), an evolutionary algorithm (Nowak-Sliwinska et al., 2016; Storn, 1996; Tsutsui et al., 2011; Wong et al., 2017), for iterative peptide generation and optimization, with HDOCK for binding affinity assessment. PISAD offers an alternative approach that does not depend on database libraries or templates, enabling the exploration of a wider sequence space. It integrates structural complementarity via molecular docking to enhance binding affinity predictions. Moreover, PISAD is designed to be computationally efficient, which supports its use in large-scale applications. To demonstrate its application, we first used PISAD for de novo peptide design targeting ADP ribosylation factor 6 (ARF6). To validate the broad applicability of PISAD, we also designed peptides targeting three additional target proteins: ADP-ribosylation factor 1 (ARF1), transforming growth factor beta 1 (TGF-β1), and interleukin-6 (IL-6).

Access through your organization

Check access to the full text by signing in through your organization.

Access through your organization

Section snippets

Peptide sequence generation

Python language was implemented to generate peptide sequences, namely initial sequence, mutated sequence, and crossover sequence. In the first iteration (I1), the algorithm commences with Z random initial sequences using the random.choices() function. This function enables the stochastic selection of amino acids from a predefined set, with each amino acid assigned a weight value that reflects its likelihood of occurrence. These weights are determined based on predefined biological preferences,

Strategy of the PISAD method

PISAD combined an iterative stochastic searching algorithm with docking assessment. The algorithmic framework of searching algorithm was designed to simulate the biological processes of peptide initial sequence generation, mutation, and crossover (Fig. 1A). Each step was designed to mimic the natural evolution of proteins, ensuring the diversity and functionality of the resulting peptide sequences.

The initial sequence generation process leverages a probabilistic model where each amino acid is

Discussion

Our study introduced PISAD, a novel iterative optimization method for de novo peptide design targeting specific proteins. PISAD integrates a differential evolution algorithm for sequence optimization, AlphaFold2 for structural modeling, and molecular docking for interaction assessment. We demonstrate PISAD's efficacy by designing peptides against ARF6 and three additional targets: ARF1, TGF-β1, and IL-6.

Traditional computational tools for designing binder peptides often rely on databases, motif

CRediT authorship contribution statement

Qiang Zhang: Writing – original draft, Investigation. Boqian Wang: Writing – review & editing. Jessica: Data curation. Behafarid Ghalandari: Methodology. Youming Chen: Investigation. Zhixiao Xu: Methodology. Quanhong Zhou: Writing – review & editing. Xianting Ding: Writing – review & editing, Supervision.

Code availability

Funding

This work was supported by: National Key R&D Program of China (2022YFC2601700). NSFC Projects (82361148715, T2122002, 22077079). Shanghai Municipal Science and Technology Project (22Z510202478). Shanghai Municipal Education Commission Project (21SG10).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

View full text

© 2025 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.