Retain
- AI optimizes mRNA vaccines by intervening during key development stages.
- It contributes to the personalization of vaccines according to the HLA profile and tumor mutations.
- It can be limited by the available data and the tumor heterogeneity that it tries to bypass.
Artificial intelligence (Ia) revolutionizes the design of mRNA vaccines, in particular in oncology and infectiology, by optimizing key stages of development, from the selection of epitopes to formulation.
AI opens the way to Personalized vaccinesadapted to the immune and genetic profile of each individual or group of patients.
How is a personalized Cancer ARNM vaccine prepared?
Vaccine manufacturing Arnm has several stages:
- Tumor biopsy and blood sample to separate tumor and normal DNA, followed by sequencing to identify specific mutations.
- Bioinformatic analysis to detect mutations and carry out the HLA typing.
- Prediction of neoantigens (tumor epitopes) by IA.
- Summary of an optimized mRNA coding 20 to 40 neoantigen.
- Encapsulation of mRNA in lipid nanoparticles (LNPS) and production (manufacturing in 4 to 8 weeks).
- Patient injection (immunological monitoring targeted by Elispot, cytometry).
What is the role of AI in the development of mRNA vaccines?
From the design, AI predicts the most immunogenic epitopes.
These peptide fragments from pathogens or tumor cells (neoantigen) are presented by the major histocompatibility (CMH) complex to T lymphocytes to trigger a targeted immune response.
To identify the best vaccination candidates who will induce a robust and sustainable immune response, the algorithms IA evaluate :
- The Affinity of CMH – Peptide link, in particular CMH List I, determining to activate CD8 cytotoxic T lymphocytes.
- Recognition by TCR receptors (TCR), an indicator of real immunogenicity.
- The processing Antigenic (antigenic fragmentation and intra-cellular transport), to ensure that the epitope is well presented.
- The effective tumor expression of neoantigens.
Integrated AI approaches like Built (Machine Unified Neoantigen Immunogenicity Scorer) et NeoaPred Use the deep learning To predict the CMH – Peptide link.
Algorithms training Based on the accumulation of data validated on neoantigens, in particular via recognized databases, such as theImmune Epitope Database (IEDB).
The AI optimizes mRNA coding the pre -selected antigenic targets.
IA algorithms intervene in:
- Optimizing the coding sequence by choosing preferential codons to maximize translation.
- Optimization of UTR non -coding regions to improve stability and efficiency.
- The modeling of the secondary structure of the mRNA, which promotes stable conformations extending in particular its half-life;
- Optimizing the global structure by reducing reasons recognized by innate immunity, limiting inflammatory effects.
For example, algorithms like Lineardesign have been developed to simultaneously optimize the structural stability of mRNA and the coding sequence.
AI participates in improving the formulation of lipid nanoparticles (LNPS), vector of mRNA.
It contributes to protection and Targeted MRNA issue in cells based on:
- The selection of lipids, in particular ionizable lipids, by predicting key physicochemical properties, such as PKA or delivery efficiency.
- The high-flow screening of formulations using LNP libraries, to adjust the proportions between the different components (cholesterol, phospholipids, peg-lipids, etc.).
The agile platformfor example, uses predictive models based in deep learning to assess the efficiency of formulations.
Are vaccines personalized by AI tested in the clinic?
The vaccine Autogen Cevumeranfor example, currently in phase II in pancreatic cancer, uses NetMHCPAN algorithm to predict the most immunogenic neoantigens.
What are the limits of AI technologies?
Although AI facilitates a accelerated design and rationalized mRNA vaccines allowing shortened clinical deadlines, it also presents limits as :
- Limited access to reliable and bulky datasets, essential to cause AI models.
- A tumor development that can cause an immune exhaust that AI can anticipate by selecting the most stable neoantigens, essential to tumor survival.
- A low generalization of predictive models to heterogeneous populations or rare cancers that AI can bypass with several predictive algorithms (overall models).
