I. INTRODUCTION
In the realm of synthetic biology and protein engineering, the quest to create optimized biological systems often comes down to refining the tools we use. This joint article introduces two such tools—Flint and ADCD—each born out of a need for more precise, accessible, and effective computational models. Developed through a combination of hard work, creativity, and collaboration, these tools seek to enhance protein receptor design and codon modification, offering practical solutions to complex biological problems.
Flint is the direct evolution of a project initially conceived during a hackathon, where the team successfully tweaked existing models to generate promising protein receptor mutants. Building on this, Flint now provides a more sophisticated and user-friendly platform for improving receptor-ligand interactions. But more on that later …
Meanwhile, ADCD offers a unique exploration into codon modifications, designed to showcase the theoretical underpinnings of PhageVO by allowing users to manipulate DNA sequences and observe potential changes. Together, these tools form a powerful duo, aiming to streamline and simplify key processes in synthetic biology, from receptor design to genetic modification.
II. AD/CD AND PHAGEVO PREDICTING
The wet team designed a wonderful tool for directed protein evolution, but we need to ensure that its users can use it without spending time and money. That's why we want to explore codon changes that would happen theoretically using Phagevo. The idea is straightforward: input a DNA sequence, and our tool will provide possible modifications using CD or AD, in forward or reverse orientations, either individually or in combination. It offers insights into adjustments in a user-friendly and accessible way ; and it's also a clear demonstration of how these modifications can work in practice.
The software is written in JavaScript, and is accessible here. The code implements several core functions that allow users to paste a DNA sequence, apply specific changes (such as CD or AD in forward or reverse orientations), and see the results. For example, translate_codon(codon) takes a codon as input and returns its corresponding amino acid based on the genetic code. The procedure apply_changes(codon, changes) applies specified changes (CD or AD in forward or reverse directions) to the codon, which means that it replaces nucleotides based on the direction and type of the change. Then, process_sequence(sequence, changes) processes an entire DNA sequence by breaking it down into individual codons. It applies the given changes to each codon using the apply_changes function and stores the results for output. These work together to analyze, apply modifications, and present the results in a clear and interactive format.
To use the ADCD tool, simply paste your DNA sequence into the provided input box on the webpage. After selecting the desired codon modification (CD or AD in forward or reverse orientation), press the button to apply the changes. The tool will display the modified DNA sequence alongside the original one, showing the differences in codon sequences and the corresponding amino acid changes in a side-by-side comparison on the webpage. The altered codons will be highlighted, helping you easily identify the specific transformations.
III. POCKETGEN AND FLINT
III.1. Hackaton
Once upon a time, even before the team's regristration at iGem, some curious developers from our college started an ambitious project : the rewriting and fine-tuning of a protein receptor design deep learning model. At this time, they had less than a week to build and present their idea. After a lot of research, hard-work, pizzas and sleepless nights, a first release was born ! Some parameters of the original model had been tweaked in order to enhance the inference results ; and it worked ! The model outputted successfully its first significatively different mutant receptor. These results conduced the team to conceive what has now became Phagevo, our team for iGem competition. After being ordered and tested by the wet lab, the mutants revealed a mid-level effectiveness, but still promising…
III.2. Deep breath
A few months later, building on the success of our early efforts at the hackathon, Flint is the next step in refining our protein receptor design model. With the lessons learned from the initial release, we’ve reimagined the approach, incorporating more sophisticated predictions for allosteric effects and designing a modular user interface to provide greater flexibility in mutation and docking settings. Flint isn’t just a continuation. It's a cleaner tool, fully rewritten and properly organized. While the initial hackathon results were promising, Flint now takes the project way further, providing greater control and deeper insights. Flint is designed to be run from the command line, allowing for easy integration into protein modeling pipelines. You can find a step-by-step guide in the repository, and a lot of documentation about how to train, install and setup the project.