Alexander D. Kalian: From LLMs to sustainable proteins: fine-tuning and prompt engineering for multi-agent AI in waste carbon-utilising microbial protein production

In this study, we present a multi-agent Artificial Intelligence (AI) framework designed to support sustainable protein production research, with an initial focus on microbial protein fermentation using side-streams which enables waste carbon capture and utilisation (CCU). Our Retrieval-Augmented Generation (RAG)-oriented system consists of two GPT-based LLM agents: (1) a literature search agent that retrieves relevant scientific literature on microbial protein production for a specified microbial strain, and (2) an information extraction agent that processes the retrieved content to extract relevant biological and chemical information. Two parallel methodologies, fine-tuning and prompt engineering, were explored for agent optimisation. 

Both methods demonstrated effectiveness at improving the performance of the information extraction agent in terms of transformer-based cosine similarity scores between obtained and ideal outputs. Mean cosine similarity scores were increased by up to 25 %, while universally reaching mean scores of ≥0.89 against ideal output text. Fine-tuning overall improved the mean scores to a greater extent (consistently of ≥0.94) compared to prompt engineering, although lower statistical uncertainties were observed with the latter approach. A user interface was developed and published for enabling the use of the multi-agent AI system, alongside preliminary exploration of additional chemical safety-based search capabilities relevant to CCU-integrated sustainable protein production.

1. Introduction

1.1. Challenges in sustainable protein production

Animal-sourced proteins are carbon-intensive and vulnerable to environmental changes. This combined with increasing protein demands highlights the complex challenge of providing protein security within sustainability targets (Piercy et al., 2023). Concurrently, rapid expansion of the manufacturing sector—particularly within the food and beverage industries—has led to the generation of large volumes of underutilised by-product streams. These side-streams, often regarded as waste, represent a valuable but underexploited opportunity for circular resource recovery and valorisation (Banks et al., 2024; Piercy et al., 2023).

Microbial proteins (MP), also known as single-cell proteins, have gained renewed attention as a promising solution to protein security. Industrial MP production dates back to the 1970s, with examples like Candida-based fermentation of volatile fatty acids (Henry et al., 1976) and Imperial Chemical Industries’ methanol-derived “Pruteen” (Braude et al., 1977). Although cost-competitive plant proteins, specifically soy—initially limited widespread adoption, the commercial success of Fusarium venenatum-based mycoprotein (marketed as Quorn™ since 1985) demonstrated the viability of MP in mainstream food markets (Whittaker et al., 2020). Grown aerobically on glucose with added nutrients, Quorn™ contains approximately 45 % protein, a complete essential amino acid profile, a healthy lipid profile, and high fibre (Banks et al., 2024; Monteyne et al., 2021). With over five billion servings to date globally, it has achieved notable health and sustainability credentials, particularly in Europe and North America (Finnigan et al., 2019).

Recent research efforts have focused on valorising nutrient-rich and contaminant-free side-streams as feedstock for production of protein alternatives using GRAS (Generally Recognised as Safe) strains (Banks et al., 2024; Piercy et al., 2023). We define such microbial proteins derived from waste carbon resources as sustainable microbial proteins. These efforts align with growing interest in waste carbon capture and utilisation (CCU), where carbon-rich industrial effluents, agro-industrial residues, and other mixed organic waste streams are repurposed as feedstock for microbial bioconversion. Such CCU strategies not only reduce carbon emissions but also enable the production of value-added bioproducts, including microbial protein.

Integrating microbial fermentation with waste carbon CCU presents a promising opportunity to valorise industrial CO₂ emissions and organic waste into high-value proteins. Our recent comprehensive review (Piercy et al., 2023) reflected the state-of-the-art research in the CCU-integrated microbial protein fermentation where the waste carbon is sourced from side streams in liquid, solid or gaseous phases (Fig. 1). Over 80 microbial species have been identified for axenic fermentation to produce food- or feed-grade proteins, with protein yields ranging from 50 to 80 wt% in bacteria, 60–70 wt% in microalgae, 30–50 wt% in fungi and yeasts, and 10–20 wt% in protists (Piercy et al., 2023).

Fig. 1. Microbial species for the CCU-integrated sustainable protein fermentation (Piercy et al., 2023).

Despite these promising yields, sustainable microbial protein production still faces critical challenges in terms of sensory properties, scalability, and economic viability. Bacterial proteins, despite high yields, often suffer from off-flavours, whereas fungal strains such as Rhizopus and Neurospora have a longer track record of use in traditional fermented foods like tempeh and oncom—early examples of “waste-to-protein” (Dalbanjan et al., 2024) - though upscale oncom products remain underdeveloped. Nevertheless, the majority of commercial MP production relies on simple, purified carbon sources (e.g., glucose, methanol), while the potential of complex and heterogeneous carbon-rich waste streams remains largely untapped. This underutilisation represents a major bottleneck for CCU-integrated sustainable protein production. Unlocking this potential will require innovations in microbial strain selection, waste carbon source screening, bioprocess optimisation, and downstream formulation to deliver microbial proteins that are not only sustainable, but also appealing in taste, texture, and cost (Lee et al., 2024).

Read full publication here