Computational blueprints expand the reach of synthetic metabolism

This review shows that modern algorithms can systematically propose synthetic routes to compounds lacking natural biosynthetic pathways, compare alternative designs, and increasingly align predictions with experimental feasibility.

Natural metabolism evolved for survival, not for producing industrial targets, which leaves many valuable molecules—such as certain diols, acids, and specialty chemicals—without native biosynthetic routes. Traditionally, building new pathways has relied heavily on expert intuition and trial-and-error experimentation, a slow and costly process. Over the past decade, advances in cheminformatics, machine learning, and systems biology have enabled computational “bioretrosynthesis,” in which algorithms work backward from a desired product to predict feasible enzymatic steps and precursor metabolites. Two broad strategies have emerged. Template-based methods use predefined reaction rules derived from known chemistry and enzymology, offering interpretability but limited coverage. Template-free approaches, often powered by deep learning, learn transformation patterns directly from large reaction datasets, expanding chemical space but raising questions about transparency and experimental realism. Despite rapid progress, the field has lacked systematic benchmarks connecting predictions to real, validated pathways.

A study (DOI: 10.1016/j.bidere.2025.100041) published in BioDesign Research on 26 July 2025 by Hongwu Ma’s team, Chinese Academy of Sciences, lays a foundation for more reliable computer-guided pathway engineering by clarifying strengths, limitations, and performance gaps across methods.

The review first details template-based strategies, explaining how reaction templates are extracted, encoded, and applied to build multi-step pathways, and how pruning algorithms and thermodynamic filters control combinatorial explosion. Case studies illustrate how such tools have successfully guided the biosynthesis of bulk chemicals, fine chemicals, and biofuels, but also expose limitations linked to rigid templates and incomplete enzyme coverage. The discussion then shifts to template-free methods, including sequence-based and graph-based neural networks that treat retrosynthesis as a translation or pattern-recognition problem. These approaches excel at discovering novel reactions and diverse pathways, yet still struggle with interpretability and experimental validation. A key contribution of the review is a curated benchmark of 55 experimentally confirmed nonnatural pathways, used to systematically evaluate representative tools from both categories. This head-to-head comparison highlights where algorithms succeed, where they fail, and why gaps persist—such as oversimplified pruning, poor handling of cofactors, and weak alignment with natural metabolic logic. Beyond comparison, the review analyzes why computational predictions often fall short in the lab. It identifies three recurring bottlenecks: overly aggressive pruning that discards viable routes, inadequate representation of large cofactors and functional groups, and limited ability to capture the sequential, natural-like organization of biological pathways. To overcome these barriers, the authors outline future directions that integrate deep learning with thermodynamics, enzyme kinetics, and genome-scale metabolic models. They argue that tighter coupling between artificial intelligence and systems-level metabolic analysis could transform pathway design from a speculative exercise into a practical engineering discipline.

In summary, this review concludes that computational design has matured into an indispensable component of nonnatural pathway engineering, but its full potential will only be realized through closer integration with experimental biology. By standardizing benchmarks, improving data representations, and fostering collaboration between computer scientists, chemists, and biologists, next-generation tools could deliver robust, user-friendly solutions for sustainable biomanufacturing and synthetic metabolism at scale.

###

References

DOI

10.1016/j.bidere.2025.100041

Original Source URL

https://doi.org/10.1016/j.bidere.2025.100041

Funding information

This work was supported by the Strategic Priority Research Program of the Chinese Academy of Sciences (XDC0110201).

About BioDesign Research

BioDesign Research is dedicated to information exchange in the interdisciplinary field of biosystems design. Its unique mission is to pave the way towards the predictable de novo design and assessment of engineered or reengineered living organisms using rational or automated methods to address global challenges in health, agriculture, and the environment.