Designing cryoprotectant agent (CPA) cocktails for vitrification is challenging because formulations must be concentrated enough to suppress ice formation yet non-toxic enough to preserve cell viability. This tradeoff creates a large, multi-objective design space in which traditional discovery is slow, often relying on expert intuition or exhaustive experimentation. We present a data-efficient framework that accelerates CPA cocktail design by combining high-throughput screening with an active-learning loop based on multi-objective Bayesian optimization. From an initial set of measured cocktails, we train probabilistic surrogate models to predict concentration and viability and quantify uncertainty across candidate formulations. We then iteratively select the next experiments by prioritizing cocktails expected to improve the Pareto front, maximizing expected Pareto improvement under uncertainty, and update the models as new assay results are collected. Wet-lab validation shows that our approach efficiently discovers cocktails that simultaneously achieve high CPA concentrations and high post-exposure viability. Relative to a naive strategy and a strong baseline, our method improves dominated hypervolume by 9.5\% and 4.5\%, respectively, while reducing the number of experiments needed to reach high-quality solutions. In complementary synthetic studies, it recovers a comparably strong set of Pareto-optimal solutions using only 30\% of the evaluations required by the prior state-of-the-art multi-objective approach, which amounts to saving approximately 10 weeks of experimental time. Because the framework assumes only a suitable assay and defined formulation space, it can be adapted to different CPA libraries, objective definitions, and cell lines to accelerate cryopreservation development.
Cryopreservation by vitrification is a cornerstone technology for long-term storage of cells, tissues, and biological systems, enabling advances across reproductive medicine, regenerative therapies, organ banking, and basic biological research [1,2]. Vitrification suppresses ice crystallization by using highly concentrated cryoprotectant agents (CPAs), thereby avoiding the mechanical and osmotic damage associated with ice formation [3]. Despite its promise, the widespread adoption of vitrification remains constrained by the toxicity of CPAs, which can severely compromise postexposure cell viability [4]. CPA toxicity is highly dependent on concentration, cell type, loading conditions, and interactions between multiple cryoprotectants, making the design of effective CPA formulations a persistent and significant challenge in cryobiology [5,6].
The prevailing strategy for mitigating CPA toxicity is to combine multiple cryoprotectants into so-called CPA cocktails, leveraging synergistic effects that reduce toxicity at a given total concentration [6,7]. Empirically, uniquely optimized CPA formulations have been developed for specific biological systems, including mammalian embryos, oocytes, organs, and insect models [8][9][10]. Experimental approaches such as median lethal dose characterization [11], kinetic toxicity modeling [12], and high-throughput screening assays [13,14] have improved our ability to quantify CPA toxicity and explore formulation spaces. However, these methods remain limited in their ability to efficiently navigate the combinatorial explosion of possible CPA cocktails, particularly when multiple objectives such as concentration and viability must be considered simultaneously.
A key limitation of existing approaches is that they rely heavily on either exhaustive experimentation or fixed experimental designs that scale poorly with the dimensionality of the formulation space. Even with automated high-throughput screening, the number of possible CPA combinations grows rapidly with the number of components and allowable concentration increments, rendering naive search strategies impractical [15]. Moreover, CPA toxicity mechanisms are only partially understood and vary across cell lines and experimental conditions, limiting the effectiveness of mechanistic or parametric models [3]. As a result, current methods struggle to balance experimental efficiency with the need to identify Pareto-optimal trade-offs between competing objectives, such as maximizing vitrification-relevant concentration while preserving cell viability.
In this work, we present a data-efficient framework for CPA cocktail optimization that integrates high-throughput screening with iterative, multi-objective Bayesian optimization. Starting from an initial set of experimentally measured CPA cocktails, we train probabilistic surrogate models that capture both predicted performance and uncertainty across the formulation space. These models are embedded within an active learning loop that selects new CPA cocktails to evaluate by maximizing expected improvement of the Pareto front under uncertainty. By explicitly accounting for both concentration and viability as competing objectives, our approach systematically balances exploration and exploitation, enabling rapid discovery of high-quality CPA formulations. Experimental validation using a validated T24 cell-based assay demonstrates that the proposed framework consistently outperforms random sampling and established scalarization-based baselines, achieving superior hypervolume and inverted generational distance metrics while requiring substantially fewer experiments.
Our main contributions are:
• a multi-objective, uncertainty-aware optimization framework for CPA cocktail design that couples high-throughput screening with Bayesian optimization,
• a batch-aware active learning strategy that efficiently explores large CPA formulation spaces while targeting Pareto-optimal trade-offs between concentration and viability,
• a generalizable methodology for accelerating cryoprotectant development that can be adapted to different CPA libraries, objective definitions, and biological systems.
In this section, we present our methodology to optimize for high-concentration, high viability CPA cocktails in an iterative manner. Figure 1 presents an overview of the optimization process, whereby experimental data is obtained through high-throughput screening of candidate CPA cocktails. This data is used to update a machine learning model that predicts the viability of any combination of the component CPAs. An optimization algorithm is used to select the next most informative batch of candidate CPA cocktails, and the process iterates. Rather than fitting a defined mathematical model to this data like Warner et al. [12], we can fit any number of regression or machine learning models to predict the viability of a candidate CPA cocktail, provided some combination of the seven component CPAs. While not
This content is AI-processed based on open access ArXiv data.