pCoMole: Pareto-Constrained Molecule Editing with Discrete Flows

Tong Chen ,

Maximilian Holsman ,

Lin Zhao ,

Pranam Chatterjee

GenAI Diffusion ML CSP Generative Models AI4S DSMSD Other AI4S

2026年10月01日

生物大分子治疗药物通常始于已知序列，需通过靶向编辑来同步提升多项性能指标，同时必须严格满足生化特性和可制造性等硬性约束条件。然而，当前的生成式方法尚无法在离散、长度可变的生物序列空间中，协同支持多目标优化、硬性可行性约束以及序列编辑这三方面需求。本研究提出“帕累托约束型分子编辑”（Pareto-Constrained Molecule Editing，简称 pCoMole）框架：该框架基于离散流匹配（discrete flow matching）构建，可在确保终态可行性的前提下，引导预训练的“编辑流”（Edit Flow）朝向用户指定的偏好方向演化。pCoMole 采用增强型切比雪夫（Tchebycheff）效用函数定义一个“可行性门控的终态分布”，并通过底层编辑过程的 Doob-h 变换，实现由此效用函数所刻画的偏好倾斜。为使该构造具备实际可操作性，我们利用对候选编辑路径进行短程蒙特卡洛 rollout 的方式，近似计算所需的调和函数（harmonic function），从而得到一种高效、可导向的编辑器，并在理论上保证其偏好一致性（preference consistency）。我们通过三类任务验证了 pCoMole 的有效性：一是缩减绿色荧光蛋白（GFP）序列长度，同时维持其荧光相关特性；二是在保持 PAM 识别特异性的前提下，缩短多种 Cas9 同源蛋白；三是将肽类结合剂压缩为短链拟肽（peptidomimetics），在多重硬性约束下同步优化七项与成药性密切相关的性质。在湿实验验证中，两个经 pCoMole 设计、仅含 229 个氨基酸残基的增强型 GFP（eGFP）变体，在 BL21 菌株中经 10 次缺失突变后，仍表现出清晰可见的绿色荧光；其中一例仅含 1 个点突变，另一例含 2 个点突变。综上，pCoMole 首次实现了在离散、长度可变的生物分子序列空间中，兼顾约束意识（constraint-aware）与帕累托最优对齐（Pareto-aligned）的序列编辑能力。

Biomolecular therapeutics often start from known sequences and require targeted editing to improve multiple properties while satisfying hard biochemical and manufacturability constraints. However, existing generative methods do not jointly support multi-objective optimization, hard feasibility, and sequence editing in discrete, variable-length biological spaces. In this work, we introduce Pareto-Constrained Molecule Editing (pCoMole), a framework built on discrete flow matching that steers a pre-trained Edit Flow toward user-specified preferences while enforcing terminal feasibility. pCoMole defines a feasibility-gated terminal distribution using an augmented Tchebycheff utility and realizes the resulting preference tilt through a Doob-h transform of the underlying edit process. To make this construction practical, we approximate the required harmonic function using short Monte Carlo rollouts over candidate edits, yielding an efficient guided editor with provable preference consistency. We validate pCoMole by shrinking GFP while retaining fluorescence-related properties, shortening diverse Cas9 orthologs while preserving PAM specificity, and compressing peptide binders into short peptidomimetics that optimize seven drug-related properties under hard constraints. In wet lab testing, two 229-residue pCoMole-designed eGFP variants retained clear green fluorescence in BL21 cells after 10 deletions, with either one or two substitutions. Together, pCoMole enables constraint-aware, Pareto-aligned editing of biomolecular sequences in discrete, variable-length spaces.

许愿