> ## Content Index
> Fetch the complete content index at: https://www.theleftshift.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Moderna’s “AI Cancer Vaccine” Is Not What You Think
- URL: https://www.theleftshift.com/modernas-ai-cancer-vaccine-is-not-what-you-think-2/
- Published: 2026-08-20T15:47:59.000Z
- Updated: 2026-08-20T15:47:59.000Z
- Description: Moderna described the system as an algorithmic and machine-learning-based process that ranks mutations according to their likelihood of producing an immunogenic response.
- Author: Pritam Bordoloi 
- Tags: AI Developments, AI News

The headline almost writes itself– [AI helped Moderna](https://www.cnbc.com/2026/08/19/moderna-merck-cancer-vaccine-shows-initial-late-stage-melanoma-data.html?ref=theleftshift.com) develop a personalised cancer vaccine, and the treatment has now succeeded in a Phase 3 trial. But that framing, while technically tempting, misses where the interesting technology actually sits.

Interestingly, the treatment is not a conventional vaccine that was designed once and given to thousands of people. Every vaccine is effectively designed for one patient. Let's dive in.

Every tumour carries its own collection of mutations. Those mutations can create abnormal proteins, known as neoantigens, that are found on cancer cells but not normally on healthy cells.

The challenge is enormous because a patient's tumour can contain thousands of genetic mutations, but only some of them will produce targets that the immune system can recognise effectively.

## Then There was Machine Learning

This is where Moderna's algorithms come in. The company takes sequencing information from a patient's tumour and normal blood cells and uses a proprietary computational system to analyse the mutations. 

The system evaluates the potential neoantigens and predicts which ones are most likely to generate an immune response. Moderna says the process can select up to 34 neoantigens for inclusion in the personalised therapy.

This is where AI played a role—but it wasn't a GPT-style large language model (LLM). Instead, it is machine learning and computational biology doing a selection job, not an LLM or AI Agent generating medical prose or inventing molecules. 

Moderna even [described](https://intismeranautogene.ai/?ref=theleftshift.com) the system as an algorithmic and machine-learning-based process that ranks mutations according to their likelihood of producing an immunogenic response. 

The company's own materials describe the current system as a **deterministic machine-learning algorithm**. An online reconstruction of the treatment pipeline similarly describes the algorithm as ranking up to 34 immunogenic neoantigens.

*"There is a (false-ish) take circling that Moderna “used AI” for this. Moderna has some marketing around it too. I’m as much of an AI for bio believer as they get, but this is not GPT-style AI. It’s more of the 2017-style modality-specific prediction model that’s worked for a while in this domain," Ankit Gupta, Y Combinator General Partner,* [*said*](https://x.com/agupta/status/2090446505568346625?ref=theleftshift.com)*.*

The company is not giving every melanoma patient the same vaccine. They are attempting to create a treatment tailored to the genetic fingerprint of an individual's tumour. 

## The AI Behind Drug Discovery Is Still More Basic Than the Hype

The use of AI in drug discovery is often portrayed as a revolution driven by sophisticated generative models, but much of its practical application remains far more basic. 

An executive from a major drug company told ***The Left Shift*** that much of the AI, particularly LLM use, in drug discovery is still related to basic tasks such as searching research papers, summarising [scientific](https://www.theleftshift.com/anthropic-expands-claudes-capabilities-to-power-scientific-discovery-in-life-sciences/) literature, analysing documents and helping researchers navigate large volumes of information. 

That makes Moderna’s use of machine learning to identify personalised cancer targets particularly interesting. Even though it represents a more specialised application of AI directly connected to the biology and design of a treatment, it's not a breakthrough as perceived.