Award-Winning Student Research

Evolving Stability

A Genetic Algorithm For Combinatorial Optimization Of Multiplexed CRISPR gRNA Array

Aryash Shyam transformed a difficult CRISPR design challenge into a mathematical optimization problem, then built a Python genetic algorithm to search 465,192 possible guide-RNA arrays for a more stable predicted structure.

Mathematical Modeling Genetic Algorithms Computational Biology Python & Bioinformatics
465,192 Possible gRNA Arrays
50 Candidates per Generation
100 Generations Evaluated
15.4% MFE Objective Improvement

Project Overview

Can an evolutionary search strategy improve the predicted stability of a multiplexed CRISPR array?

Evolving Stability is an in silico research project by Aryash Shyam that applies a genetic algorithm to the optimization of multiplexed CRISPR guide-RNA arrays. Aryash modeled 465,192 possible combinations targeting KRAS, TP53, and EGFR, evaluated predicted RNA folding with ViennaRNA RNAfold, and improved the project’s Minimum Free Energy objective from −112.70 kcal/mol to −130.10 kcal/mol.

The Challenge

Why multiplexed CRISPR creates a mathematical design problem

Multiplexed CRISPR allows researchers to target several genes at once by placing multiple guide RNA sequences into a single RNA array. As the molecule becomes longer, however, it can fold into stems and loops that may interfere with processing and performance.

01

The biological constraint

Each guide RNA may be promising on its own, yet its behavior can change when combined with other sequences. The full array must therefore be evaluated as one interacting RNA structure rather than as a collection of independent parts.

02

The mathematical objective

Aryash used Minimum Free Energy, or MFE, as the objective function. In this computational model, a lower and more negative MFE represents a more thermodynamically stable predicted RNA structure.

Mathematical Formalization

Turning CRISPR array design into a searchable solution space

Aryash assembled three candidate guide-RNA pools: 63 sequences for KRAS, 104 for TP53, and 71 for EGFR. Selecting one guide RNA from each pool produces an ordered three-part array.

Candidate solution

Each possible array was encoded as an ordered vector:

A = (g KRAS , g TP53 , g EGFR ) One guide RNA selected from each gene pool

Search-space size

The complete space is the Cartesian product of the three candidate pools:

63 × 104 × 71 = 465,192 Distinct ordered combinations

Method

How Aryash built the genetic algorithm

The Python model evolved a population of candidate arrays over 100 generations. It combined the strongest features of promising solutions while preserving enough randomness to explore new regions of the search space.

1

Initialize

Create a population of 50 randomly assembled gRNA arrays.

2

Evaluate

Use ViennaRNA RNAfold to calculate the predicted MFE of each array.

3

Select

Run tournaments of three candidates and favor the array with the lower MFE.

4

Recombine

Apply single-point crossover and a 20% mutation rate to create new candidates.

5

Preserve

Carry the best solution forward through elitism so the leading result cannot worsen.

Key Result

A stronger predicted RNA structure emerged through iterative optimization

The largest gains occurred during the early generations, followed by a gradual plateau as the population converged.

Initial Random Best −112.70 kcal/mol
GA-Optimized Result −130.10 kcal/mol
15.4%

improvement in the project’s MFE objective. ViennaRNA secondary-structure predictions also showed a greater concentration of high-probability base-pairing regions in the optimized array. These findings remain computational predictions rather than laboratory validation.

Skills Demonstrated

Research at the intersection of mathematics, coding, and biology

Mathematical Modeling Defined a Cartesian-product search space and an objective-function minimization problem.
Algorithm Design Selected and implemented tournament selection, crossover, mutation, and elitism.
Python Programming Built the genetic algorithm, data pipeline, RNAfold interface, testing tools, and plots.
Bioinformatics Worked with gRNA pools, RNA secondary-structure prediction, and MFE-based evaluation.
Data Analysis Tracked best fitness, population mean, standard deviation, diversity, and convergence.
Scientific Communication Translated advanced CRISPR and optimization concepts into clear explanations and visual models.
A

Compared search strategies

Evaluated hill climbing and simulated annealing before selecting a population-based genetic algorithm.

D

Built the data foundation

Compiled and deduplicated candidate guide-RNA sequences for KRAS, TP53, and EGFR using CRISPOR and Addgene.

B

Diagnosed logic errors

Corrected an offspring-generation bug, a maximum-versus-minimum selection error, and population stagnation.

V

Improved validation

Used fixed random seeds, small test cases, assertions, unit tests, average fitness, and standard deviation to examine model behavior.

Limitations & Future Work

What the model can show—and what it cannot yet prove

Current limitation

The model optimizes one computational objective: Minimum Free Energy. MFE helps compare predicted RNA structures, but it does not capture every factor that determines performance inside a living cell. Evolving Stability is therefore an in silico study, not experimental or clinical validation.

Multi-objective optimization Balance MFE with on-target efficiency, off-target risk, and sequence constraints.
Algorithm benchmarking Compare the genetic algorithm with simulated annealing and other heuristic search methods.
Design-build-test-learn Connect computational predictions with laboratory measurements and use the results to refine the model.

Competition Recognition

Recognized across regional and state science competitions

The awards honored the project’s combination of mathematical rigor, computational problem-solving, biological context, and clear scientific presentation.

Lehigh Valley Science Fair First Place
Delaware Valley Science Fairs Medal Recognition
PJAS Regionals First Place
PJAS State Competition First Place · Perfect Score

Frequently Asked Questions

About Evolving Stability

What is Evolving Stability?

Evolving Stability is an award-winning research project by Aryash Shyam that applies a Python genetic algorithm to the mathematical optimization of multiplexed CRISPR guide-RNA arrays.

How large was the search space?

The model considered 465,192 possible ordered arrays created from 63 KRAS guide RNAs, 104 TP53 guide RNAs, and 71 EGFR guide RNAs.

How did Aryash evaluate RNA stability?

Aryash used Minimum Free Energy as the objective function and calculated predicted RNA secondary structures with ViennaRNA RNAfold. A lower, more negative MFE represented greater predicted thermodynamic stability in the model.

What result did the genetic algorithm achieve?

The best initial random array had an MFE of −112.70 kcal/mol, while the optimized array reached −130.10 kcal/mol, a 15.4% improvement in the project’s MFE objective.

Was the project tested in a laboratory?

No. Evolving Stability is an in silico computational study. Its results are predicted RNA-structure and energy values, not laboratory validation of gene-editing performance.

Which competitions recognized the project?

Evolving Stability received First Place at the Lehigh Valley Science Fair, medal recognition at the Delaware Valley Science Fairs, First Place at PJAS Regionals, and First Place with a perfect score at the PJAS State Competition.

The Bigger Idea

From trial-and-error selection to a formal computational search

Evolving Stability demonstrates how combinatorics, probability, optimization, programming, and RNA thermodynamics can work together to narrow a complex biological design space before costly laboratory experimentation begins.

Research & Accolades