Serum-Free Cell Culture Media Design Using an Artificial Intelligence Algorithm

BACKGROUND

Serum-free cell culture media design is important because it offers the opportunity to construct chemically defined media that possess greater batch-to-batch reproducibility and clinically compatible reagents; benefits not necessarily associated with serum-containing media.  As such, it is an ideal method for producing media for the growing cell therapy market segment where high quality and reproducibility standards influence clinical translation.  However, the use of defined chemical factors often creates a complex, large-scale optimization problem where the number of possible combinations of factors across different dose concentrations creates a search space so large that only a millionth or less of it can be sampled.  Consequently, this renders traditional media optimization techniques ineffective and necessitates the development of new techniques to handle this problem.

 

TECHNOLOGY

Researchers at the University of Toronto had previously constructed a high dimensional-differential evolution (HDDE) algorithm to optimize media containing as many as 15 different chemical factors.  They used it to produce serum-free media for the expansion of both TF-1 and CAR-T cells that is comparable to gold standard media.  More recently, they have added an artificial neural network module to the HDDE algorithm (HiDiNeu) enabling both faster determination of optimal media composition and optimization of exponentially larger search spaces (Figure 1).  This module was tested in silico using standard benchmark functions where it was shown to reach a close-to-optimum formulation in only two generations (3x quicker than the HDDE algorithm) and with higher fidelity to the true solution.  It’s also important to note that such algorithms can optimize any measurable cell parameter beyond cell growth and maintenance (e.g. contractility, differentiation potential, etc.).

Figure 1.  Serum-free cell media optimization using algorithms.  For a given number of factors (A, B, C, . . f) and dose concentrations (coloured squares), a small subset of factor combinations (compared to the total search space) is generated for experimental testing.  From these experimental results, a model of the search space is generated using an automated AI algorithm (i.e. requiring no user input) to assess optimization.  A decision is then taken to either suggest further experiments to more accurate define the search space or, if an optimal media constitution is identified, the loop is halted with the optimized media solution proposed.

 

COMPETITIVE ADVANTAGE

  • Serum-media optimization achieved with miniscule sampling of the solution space
    • <5x10-6 % of the total search space
  • Optimization of large number of factors possible
    • ~ 25 chemical factors or greater
  • AI module increases speed, reduces work, and increases accuracy of solution
    • 2x faster
    • 2x less experimental work needed
    • >50% reduction in error of solution (compared to non-AI algorithms)
    • Highly robust to data variance

 

APPLICATIONS

  • Serum-free media design and optimization
  • Cell manufacturing processes

 

INTELLECTUAL PROPERTY STATUS

  • Provisional patent application filed (Dec 2021)

 

PROJECT STATUS

The high-dimensional differential evolution algorithm (HDDE) was originally tested in silico, after which it was used to produce serum-free media for the expansion of both TF-1 and CAR-T cells that was comparable to gold standard media.  The HDDE algorithm incorporating an AI module (HiDiNeu) has been tested in silico on three high dimensional benchmark functions (Rosenbrock, Ackley, Rastrigen).

ID:

P2302

Keywords:

, Research Tools

Related Resources

VPRI Contact

Jae-Ho Yoo

Innovations and Entrepreneurship Manager
Innovations & Partnerships Office (IPO)
416-946-7158