AI-Driven Enzyme Design for Advanced Plastic Recycling

rhea’s_factory · Energy · 2026 · source Agentic: no · promotional bias: med · source-backed · extracted with prompt 0.5.4 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Enzyme design for plastic recycling

Rhea's Factory needed to design enzymes that can break plastic polymers back into monomers under practical low-temperature, low-pressure recycling conditions.

Constraints:

wet-lab experiment cost
large protein design space
selectivity across mixed plastic waste
low-temperature and low-pressure process goals

Architecture

Rhea's Factory uses computational enzyme-design modules and protein-language-model style prediction to explore enzyme candidates, then sends a small selected subset into wet-lab validation for plastic-recycling performance.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  computational-design-modules["Computational design modules"]
  protein-language-models["Protein language models"]
  predictive-screening["Predictive screening"]
  wet-lab-validation["Wet-lab validation"]
  candidate-feedback-loop["Candidate feedback loop"]
  computational-design-modules --> protein-language-models
  computational-design-modules --> predictive-screening
  protein-language-models --> predictive-screening
  predictive-screening --> wet-lab-validation
  wet-lab-validation --> candidate-feedback-loop
  candidate-feedback-loop --> computational-design-modules

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Computational design modules`	Explore enzyme variants and protein design space before lab work.	`Protein language models`, `Predictive screening`
`Protein language models`	Model protein sequences and structures for candidate enzyme design.	`Computational design modules`, `Predictive screening`
`Predictive screening`	Ranks or filters designed enzymes so only a subset goes to wet-lab testing.	`Wet-lab validation`
`Wet-lab validation`	Tests selected enzyme candidates against recycling targets.	`Candidate feedback loop`
`Candidate feedback loop`	Uses lab results and domain judgment to refine future computational design.	`Computational design modules`

Data flow

The workflow moves most enzyme exploration into machine-learning design, narrows candidates computationally, and then validates only a subset in the lab.

“we can test only a subset a small number in our lab which accelerates our timelines and also decreases the resources that we use.”transcript, stated

Why this architecture

This architecture exists because trial-and-error enzyme engineering requires physically building candidates and is too slow for broad plastic-recycling targets.

“the traditional methods which is mainly a trial and error. >> Yeah. >> You just test things in the lab. You come up with an idea. you design it and you have to physically build this enzyme”transcript, stated

Models used: protein language models · machine learning models

Scale

The recycling process is designed for mild conditions around 60 Celsius rather than high-heat processing.
“However, they are biological organisms. Therefore, they live and operate in a very mild conditions compared to 1,00 Celsius degree heat that we process. We are like operating at 60 Celsius for example.”transcript, stated

Alternatives considered (roads not taken)

Pure wet-lab trial and error

Rejected because: The team moved discovery upstream into computation so lab work becomes targeted validation instead of broad search.

Natural enzyme search only

Rejected because: The team designs enzymes that nature has not seen before instead of limiting the search to known biological variants.

“Now we tackle most of this problem computationally using the machine learning tools and design new enzymes that the nature hasn't seen before.”transcript, stated

Maximizing lab experiments

Rejected because: The goal is fewer, better-targeted wet-lab tests with a higher hit rate rather than more brute-force experiments.

“Are you running are you able to run more wet lab experiments? Is the goal to run fewer and have a higher hit rate?”transcript, implied

Patterns

`human-in-the-loop-gate` (scientific-workflow)

The platform computationally proposes enzyme candidates, but wet-lab testing remains the gate before any candidate is treated as real.

“Now we tackle most of this problem computationally using the machine learning tools and design new enzymes that the nature hasn't seen before.”transcript, stated

`data-quality-loop` (scientific-workflow)

The team uses lab results to reduce the number of candidates tested while increasing diversity and expected hit rate.

“we can test only a subset a small number in our lab which accelerates our timelines and also decreases the resources that we use.”transcript, stated

Tradeoffs

How much of enzyme discovery to move into computation.

Chose AI-designed candidates with limited wet-lab validation over pure trial-and-error laboratory exploration.

Rationale: Computation expands design diversity and reduces resource use, but the wet lab remains necessary for proof.

“What's been the impact of having an AI scientist on your team? The impact is again on the definitely the timelines and also the diversity and the design space we can get in”transcript, stated

How much experimental search to outsource to protein models

Chose computational candidate generation plus targeted wet-lab validation over broad physical construction and testing of enzyme variants.

Rationale: Protein models expand candidate diversity while lab work remains the final filter for practical recycling performance.

“machine learning with the algorithms can design this new enzymes and can present them to us and we can test only a subset a small number in our lab”transcript, stated

Gotchas (what bit them)

Traditional enzyme engineering is too slow and resource-intensive for broad plastic targets.

Root cause: Trial-and-error design requires physically building and testing many enzyme variants in the lab.

Resolution: Use computational design to narrow the candidate set before wet-lab validation.

Evaluation

The case evaluates enzyme design by comparing computational candidate generation against wet-lab testing needs, design diversity, and process constraints.

Metrics: timeline acceleration · design-space diversity · wet-lab experiment reduction · low temperature and pressure feasibility