Source-linked AI summary
Genesis: Towards the Automation of Systems Biology Research
Ievgeniia A. Tiukova, Daniel Brunnsåker, Erik Y. Bjurström, Alexander H. Gower, Filip Kronström, Gabriel K. Reder, Ronald S. Reiserer, Konstantin Korovin, Larisa B. Soldatova, John P. Wikswo, Ross D. King
TL;DR
Systems biology models contain many interacting components, while current high-throughput experiments are not hypothesis-led. Genesis combines automated model improvement and experimental cycles with supporting software and laboratory systems; its components have demonstrated model, database, and hypothesis-generation capabilities, though the system is not yet at full capacity.
Problem
Biological systems contain thousands of interacting components, and current high-throughput experiments are not designed to test model hypotheses.
Source: §1.2 · p. 2Method
Genesis combines hypothesis-led closed-loop experiments with AutonoMS for mass-spectrometry workflows and Genesis-DB and RIMBO for structured information and model revisions.
Source: §2 · p. 3, §2.2 · p. 5, §2.4 · p. 7Results
LGEM+ proposed 2,094 unique candidate hypotheses, and deductive inference on its logical theories predicted yeast-strain growth or no-growth in minimal media.
Source: §2.6 · p. 10Takeaways & Limitations
Genesis-DB has been used to model yeast gene regulation and guide future hypothesis generation and experiment design.
Source: §2.3 · p. 6Genesis’s core structure is complete, but hardware and software work remains before it operates at full capacity.
Source: §3.1 · p. 10
Abstract
from arXiv · showhide
The cutting edge of applying AI to science is the closed-loop automation of scientific research: robot scientists. We have previously developed two robot scientists: `Adam' (for yeast functional biology), and `Eve' (for early-stage drug design)). We are now developing a next generation robot scientist Genesis. With Genesis we aim to demonstrate that an area of science can be investigated using robot scientists unambiguously faster, and at lower cost, than with human scientists. Here we report progress on the Genesis project. Genesis is designed to automatically improve system biology models with thousands of interacting causal components. When complete Genesis will be able to initiate and execute in parallel one thousand hypothesis-led closed-loop cycles of experiment per-day. Here we describe the core Genesis hardware: the one thousand computer-controlled $\mu$-bioreactors. For the integrated Mass Spectrometry platform we have developed AutonoMS, a system to automatically run, process, and analyse high-throughput experiments. We have also developed Genesis-DB, a database system designed to enable software agents access to large quantities of structured domain information. We have developed RIMBO (Revisions for Improvements of Models in Biology Ontology) to describe the planned hundreds of thousands of changes to the models. We have demonstrated the utility of this infrastructure by developed two relational learning bioinformatic projects. Finally, we describe LGEM+ a relational learning system for the automated abductive improvement of genome-scale metabolic models.
1 Background
Robot Scientists close the loop from hypothesis formation through robotic experiments to interpretation, while systems biology faces models too complex for human intuition and current high-throughput methods that are not hypothesis-led. The earlier yeast diauxic-shift work demonstrates automated model improvement, adding genes and interactions.
1.1 Automating Science
- Robot Scientists autonomously form hypotheses, devise experiments, run them with laboratory robotics, interpret results, and repeat the cycle.The automation of science has the potential to improve research efficiency.Source: §1.1 · pp. 1–2
- Adam investigated yeast functional genomics and discovered locally orphan enzyme functions, while Eve was designed for early-stage drug development.Eve’s AI-selected compounds outperformed standard screening, and its closed-loop optimisation approach was widely copied in pharmaceutical and materials science.Source: §1.1 · p. 2
1.2 Eukaryotic System Biology
- Systems biology models of eukaryotic cells are challenging to develop, and yeast is a model organism whose biology can inform study of human genes through homologs.Humans and yeast share similarities despite their last common ancestor living around one billion years ago.Source: §1.2 · p. 2
- Biological systems involve thousands of interacting components, and current high-throughput experiments are not hypothesis-led; model complexity and the Duhem Thesis make efficient testing and refinement especially challenging.The authors argue AI tools are required to support the many closed-loop experimental cycles needed to build accurate and comprehensive systems biology models.Source: §1.2 · p. 2
- The diauxic-shift model outperformed previous models and added 92 genes (+45%) and 1,048 interactions (+147%).The resulting model is relevant to understanding cancer, the immune system, and aging.Source: §1.2 · p. 3
2 Results
Genesis is being built to automate high-throughput systems-biology model improvement, pairing large-scale experimental hardware with software and data infrastructure. This section reports progress in mass-spectrometry automation, ontologies, model revision, bioinformatics, and logical-model learning.
2 Results
- Genesis targets one thousand parallel hypothesis-led experiment cycles per day for models with thousands of interacting causal components, rather than black-box input/output optimization.The project aims to demonstrate a hundredfold cost-benefit over a standard human scientist in a standard lab.Source: §2 · p. 3
2.1 Genesis Hardware
- The planned 1,000 µ-bioreactors, arranged in 48-unit groups, support batch, fed-batch, or continuous experiments with growth, metabolite, and gene-expression readouts.The planned observables include ∼100 internal metabolites and comprehensive expression levels for ∼6,000 genes.Source: §2.1 · p. 4
2.2 Mass Spectrometry
- 5 second sample analysis was demonstrated on robotic standards and cultured yeast, while planned metabolomics capacity is ∼10,000 measurements a day.The demonstrated flow-injection ion-mobility setup covered targeted and untargeted metabolomics applications.Source: §2.2 · p. 5
2.3 Ontologies
- Genesis-DB demonstrated a research cycle using database metadata to reconstruct gene regulation and guide future hypotheses and experiment design.The illustrated cycle incorporates a new high-temperature experiment into the reconstructed regulatory network.Source: §2.3 · p. 6, Figure 4 · p. 6
2.4 Model Revision
- Thousands of simulated revisions and the Yeast8 8.4.1-to-8.4.2 update were represented semantically meaningfully and storage-efficiently in a RIMBO-based database.RIMBO is intended to record revision chains, including the reasons and descriptions for changes.Source: §2.4 · pp. 6–7
2.5 Bioinformatics
- Ten regulatory deletant strains were profiled with untargeted metabolomics to investigate deletion consequences and regulatory roles during the diauxic shift.The study used prior semi-autonomously developed gene-regulatory models to select the ten genes.Source: §2.5 · p. 7
- Relationally learned descriptors from yeast databases and sample metadata supported explainable protein-abundance prediction, linking abundance with function and phenotype.Examples connected α-amino acid accumulation and chronological-lifespan deviations to protein abundance, including His4 and Ilv2.Source: §2.5 · p. 8, Figure 6 · p. 8
2.6 Learning Logical Models
- 2,094 unique candidate hypotheses for model improvement were generated by LGEM+, whose logical theories also predicted minimal-media growth/no-growth for S. cerevisiae strains.LGEM+ combines a compartmentalised first-order-logic framework based on curated GEMs with two-stage hypothesis abduction.Source: §2.6 · pp. 9–10
3 Discussion
Genesis has its core structure, but significant hardware and software work remains before it reaches full capacity. The discussion also considers LLMs for formalizing scientific knowledge and forming hypotheses, including hypotheses whose validity can be tested experimentally.
3.1 Completing Genesis
- Genesis’s core structure is complete, but hardware integration, software development, and control systems remain before it operates at full capacity.Work includes scaling and integrating the µ-bioreactors, integrating mass spectrometry and planned RNA-SEQ, developing agent and learning systems, and implementing robust control and a virtual Research Agency.Source: §3.1 · p. 10
3.2 Large Language Models
- LLMs could convert natural-language scientific knowledge into formal knowledge that can be checked and reasoned with, though their most important impact on Genesis remains unclear.The paper also states that LLMs solve the predicate invention problem.Source: §3.2 · p. 10
- LLM-generated hypotheses may include hallucinations, which can be useful when their validity can be objectively tested in laboratory experiments.The paper distinguishes factually wrong outputs from hypotheses whose validity is uncertain.Source: §3.2 · p. 10