Work overview

Section 06 of 06

6 Discussion

seqme: a Python library for evaluating biological sequence design from generative models

Rasmus Møller-Larsen, Adam Izdebski, Jan Olszewski, Pankhil Gawade, Michal Kmicikiewicz, Wojciech Zarzecki, and Ewa Szczurek · 2026

Contents

Section 06 of 06

  1. 011 Introduction
  2. 022 Metrics
  3. 033 Additional functionalities of seqme
  4. 044 Implementation
  5. 055 Case studies
  6. 066 Discussion
Text size
Work overview

Section 6 of 6

6 Discussion

Rasmus Møller-Larsen, Adam Izdebski, Jan Olszewski, Pankhil Gawade, Michal Kmicikiewicz, Wojciech Zarzecki, and Ewa Szczurek · about 1 minutes

Original source ↗

seqme aims to ease the task of evaluating biological sequence design methods by providing metrics, as well as embedding and property models. While the library is already applicable to multiple biological sequence types, it is easily extendable to more types by adding type-specific embedding and property models.

We encourage users to use multiple metrics provided by seqme, as each metric offers a different perspective. This is of particular importance as each evaluation metric has limitations and can fail to detect a failure mode in a generative AI model or algorithm used for the design (Räisä et al. 2025). Only a well-chosen set of metrics will yield a robust evaluation.

We envision seqme will accelerate de novo drug discovery by providing the foundation for robust model benchmarking, and offering more comprehensive tools for defining stopping criteria in machine-learning training loops. Ultimately, we expect that libraries like seqme will enable practical applications of generative AI and other sequence design methods, and facilitate the translation of computational advances into biology and medicine.