Researchers have introduced a new Python package called PoolParty, designed to streamline the computational design of DNA sequence libraries for high-throughput biological experiments. Published in BMC Bioinformatics, the software addresses a persistent challenge in functional genomics: the tedious and error-prone process of manually scripting complex oligonucleotide pools. By representing library designs as directed acyclic graphs, PoolParty allows scientists to define intricate sequence generation logic using a simple and flexible application programming interface. This approach transforms a task that typically requires custom, one-off scripts into a structured, transparent, and reproducible process, significantly reducing the potential for human error in library construction.
The development of PoolParty responds to the increasing reliance on computationally designed DNA libraries in massively parallel reporter assays, deep mutational scanning, and other multiplex assays of variant effect. These libraries are essential for measuring regulatory activities or mapping protein fitness landscapes at scale. However, designing them often involves coordinating multiple types of variant sequences, each generated by different procedures and subject to distinct coverage requirements. For instance, a deep mutational scanning library might combine exhaustive single-amino-acid substitutions with sampled higher-order mutants, wild-type controls, and barcodes. Coordinating this combinatorial logic in a single script quickly becomes unwieldy, often producing errors that are difficult to detect without specialized tools.
Existing tools such as VaLiAnT, MPRAnator, and tangermeme offer partial solutions but are typically targeted to specific applications. VaLiAnT generates mutational scanning libraries for coding and noncoding regions, while MPRAnator provides modules for motif placement and SNP design in reporter assays. However, no single tool previously supported the full range of library designs in common use. For example, VaLiAnT does not support regulatory sequence libraries containing random combinations of transcription factor binding sites, and other tools lack support for codon-aware mutagenesis of protein-coding sequences. PoolParty bridges this gap by providing a unified framework that can handle diverse library types, from deep mutational scanning to regulatory grammar studies.
The core architecture of PoolParty is built around two abstractions: Pools and Operations. Pools represent collections of sequences, which can be the final library or intermediate sets used in the construction process. Operations represent the specific steps used to create these Pools, such as mutagenesis, motif insertion, or barcode generation. Users chain these Operations together into a directed acyclic graph, an approach inspired by deep learning frameworks. This graph describes the high-level logic of sequence generation, while the procedural details and bookkeeping are handled internally by the software. A key feature of this design is that no sequences are generated until the user explicitly requests them, allowing researchers to explore multiple design options and test small subsets of sequences without committing to generating the entire library at each iteration.
PoolParty includes over 50 built-in Operations that cover nucleotide- and codon-level mutagenesis, motif insertion, and barcode generation. These Operations fall into four functional categories: source, transformation, composition, and state. Source Operations create initial Pools from user-specified sequences, position weight matrices, or random k-mers. Transformation Operations modify sequence content through point mutations, insertions, deletions, or recombination. Composition Operations combine sequences from multiple parent Pools, either by concatenating them end-to-end or by merging multiple Pools into one. State Operations affect which sequences are kept and how they are ordered, including methods to select, reorder, filter, and replicate sequences based on criteria such as GC content, homopolymer length, or restriction enzyme recognition sites.
To ensure reproducibility and transparency, PoolParty automatically generates informative names for each sequence and provides “design cards” that detail how each sequence was constructed. These design cards record the specific changes applied to each sequence by each Operation, including user-defined sequence scores. This metadata is crucial for downstream analyses, particularly in in silico experiments where the parameters governing sequence construction can be used directly as covariates. For example, in a surrogate modeling analysis of the SpliceAI genomic deep learning model, the design cards provided the covariates needed to fit generalized additive models to the model’s predictions. This capability allows researchers to systematically probe and interpret the behavior of genomic AI models without reverse-engineering sequence architecture from raw data.
The authors demonstrated the utility of PoolParty through three example applications. The first was a deep mutational scanning library for the protein GB1, which reconstructed and extended a benchmark dataset containing over half a million variants. Using PoolParty, the researchers generated a library of 547,230 sequences, including wild-type replicates, all single-amino-acid substitutions, all pairwise amino acid substitutions, and random higher-order variants. The design cards for each variant reported the library component, specific mutations, and replicate index, allowing for easy inspection of the library’s structure. The second example was a massively parallel reporter assay library designed to probe transcriptional regulatory grammar. This library varied the positions and orientations of binding sites for three liver-enriched transcription factors, generating 24,000 unique sequences. The software’s visualization methods allowed users to quickly audit the library content and inspect the underlying graph, with colors indicating mutations, deletions, and specific regions of interest.
The third example involved an in silico experiment characterizing the SpliceAI model, which predicts splice-site usage in humans. The researchers used PoolParty to construct a library of cryptic 5’ splice site sequences of varying strength and position, inserted near the canonical 5’ splice site of SMN2 exon 7. Each variant sequence was paired with a matched control in which the splice-site GT dinucleotide was disrupted. The effect of each insertion was quantified as the difference in logit-transformed SpliceAI-predicted probabilities at the canonical site. Generalized additive models were fitted to the data, revealing that cryptic sites on the exonic side suppressed canonical site usage more strongly than those on the intronic side, with the effect peaking approximately 20 base pairs upstream of the canonical site. The models also showed substantial interactions between position and strength, with held-out R-squared values of 0.82 and 0.71 for exonic and intronic insertions, respectively.
PoolParty is designed to be extensible, allowing users to create new Operation types by subclassing the Operation base class. Custom Operations automatically inherit support for state tracking, design card generation, and visualization, enabling researchers to add domain-specific functionality without reimplementing core infrastructure. The package requires Python 3.10 or later and is distributed via PyPI, with documentation and tutorials available on ReadTheDocs and GitHub. The authors note that PoolParty complements existing tools by providing a single framework for diverse library designs, but it has limitations. It operates on sequences rather than genomic loci or transcript models, meaning it does not interpret transcript annotations or track genomic coordinates through design steps. It is also not a sequence optimization tool and does not design sequences with optimized codon usage or subject to complex synthesis constraints.
Despite these limitations, PoolParty represents a significant advance in the computational tools available for functional genomics. By streamlining the design of DNA sequence libraries, it reduces the barrier to entry for complex multiplex assays and facilitates the systematic probing of genomic AI models. As these models continue to advance, the ability to generate and analyze large, well-characterized sequence libraries will become increasingly important. PoolParty’s composable design and automated metadata generation position it as a valuable resource for researchers seeking to explore the sequence-function relationship at scale. The software is available under the MIT License, ensuring broad accessibility for both academic and non-academic users. The authors affirm that they are fully responsible for the content of the manuscript, the codebase, and the documentation, with large language models used only for copy-editing and plot script generation.
Subject of Research: Computational design of DNA sequence libraries for functional genomics
Article Title: PoolParty: streamlined design of DNA sequence libraries in Python
Article References: Liu, Z., Cordero, A., & Kinney, J. B. (2026). PoolParty: streamlined design of DNA sequence libraries in Python. BMC Bioinformatics, 27(1), Article 253. https://doi.org/10.1186/s12859-026-06696-6
Image Credits: AI Generated
DOI: 10.1186/s12859-026-06696-6
Keywords: PoolParty, DNA sequence libraries, Python package, Deep mutational scanning, Massively parallel reporter assays, Genomic AI, Bioinformatics, Directed acyclic graphs, Oligonucleotide pools, Functional genomics, streamlined, design
News Source: Gregory Coleman. (November 8, 2026). Python Package Simplifies Complex DNA Library Design. Scienmag.
