Sign in Subscribe
5 min read

Agents start naming drug targets

Share
Agents start naming drug targets
Nº 01 · The Lede bioRxiv Spatial transcriptomics

Spatial transcriptomics agent flags ICAM-1 in lung fibrosis

Spatial transcriptomics agent flags ICAM-1 in lung fibrosis
Fig. IbioRxiv · Filed 27 Oct 2026.

An AI agent prioritized ICAM-1 as a target in lung fibrosis after analyzing spatial transcriptomics data, according to a new bioRxiv preprint. The agent is software that runs multi-step analyses on its own, and it works inside a control layer the authors call "evidence-governed." That label is aimed at the main worry about letting AI pick targets, which is that a model can return a confident ranking with nothing behind it. The name signals a design meant to tie each call to supporting evidence. Spatial data records where in tissue a gene is expressed, not only how much, and the agent reasoned over that data to reach ICAM-1.

Read the source →

AI agents derive bioprocess kinetic models on their own
Fig. IIarXiv · Filed 27 Oct 2026.
Nº 02 arXiv Agents · Infrastructure

AI agents derive bioprocess kinetic models on their own

LLM-driven agents worked out the structure of bioprocess models without step-by-step human direction, according to a new arXiv preprint covering both process dynamics and kinetics. LLMs (large language models, the systems behind chatbots such as ChatGPT and Claude) act here as agents that propose candidate model structures for how a bioprocess behaves over time. Picking that structure is usually expert work, with people proposing, fitting and comparing models by hand. In the preprint, the agents do that search instead.

Read more →
Preprint proposes self-improving clinical AI without a hub
Fig. IIIarXiv · Filed 27 Oct 2026.
Nº 03 arXiv Clinical AI · Evaluation

Preprint proposes self-improving clinical AI without a hub

An arXiv preprint proposes decentralized clinical agents that keep improving themselves, and argues they are a step past federated learning. Federated learning trains one shared model across hospitals by sending model updates, not patient records, to a central server. The preprint drops that central server in favor of a decentralized network of agents, which the authors pitch as "self-evolving clinical intelligence." Self-evolving means the system is meant to keep updating as it runs instead of staying fixed after training.

Read more →
Also Filed · One Brief from the queue
Nº 04 bioRxiv Field report

Synthetic data recovers disease signatures in small studies

Model-generated samples recover prognostic disease signatures from transcriptomic datasets too underpowered to show them, according to a bioRxiv preprint. The approach, called generative augmentation, pads a small dataset with synthetic samples produced by a generative model.

Read →

Reply with your discoveries. A human reads them. Forward freely.

Agentic Discovery  ·  Nº 129  ·  27 Oct 2026

Editor's Note

Today's agents name a fibrosis target, derive kinetic models and ditch the central server.

 

Nº 01 · The Lede  —  bioRxiv  —  Spatial transcriptomics

Spatial transcriptomics agent flags ICAM-1 in lung fibrosis

Spatial transcriptomics agent flags ICAM-1 in lung fibrosis

Fig. I  bioRxiv · Filed 27 Oct 2026.

An AI agent prioritized ICAM-1 as a target in lung fibrosis after analyzing spatial transcriptomics data, according to a new bioRxiv preprint. The agent is software that runs multi-step analyses on its own, and it works inside a control layer the authors call "evidence-governed." That label is aimed at the main worry about letting AI pick targets, which is that a model can return a confident ranking with nothing behind it. The name signals a design meant to tie each call to supporting evidence. Spatial data records where in tissue a gene is expressed, not only how much, and the agent reasoned over that data to reach ICAM-1.

Read the source →

Why it matters

It puts a named, testable target on the table, so claims about agents doing target discovery can be checked at the bench rather than taken on trust.

 

Nº 02  —  arXiv  —  Agents · Infrastructure

AI agents derive bioprocess kinetic models on their own

Fig. II  arXiv · Filed 27 Oct 2026.

AI agents derive bioprocess kinetic models on their own

LLM-driven agents worked out the structure of bioprocess models without step-by-step human direction, according to a new arXiv preprint covering both process dynamics and kinetics. LLMs (large language models, the systems behind chatbots such as ChatGPT and Claude) act here as agents that propose candidate model structures for how a bioprocess behaves over time. Picking that structure is usually expert work, with people proposing, fitting and comparing models by hand. In the preprint, the agents do that search instead.

Read more →

 

Nº 03  —  arXiv  —  Clinical AI · Evaluation

Preprint proposes self-improving clinical AI without a hub

Fig. III  arXiv · Filed 27 Oct 2026.

Preprint proposes self-improving clinical AI without a hub

An arXiv preprint proposes decentralized clinical agents that keep improving themselves, and argues they are a step past federated learning. Federated learning trains one shared model across hospitals by sending model updates, not patient records, to a central server. The preprint drops that central server in favor of a decentralized network of agents, which the authors pitch as "self-evolving clinical intelligence." Self-evolving means the system is meant to keep updating as it runs instead of staying fixed after training.

Read more →

 

Also Filed  ·  One Brief from the queue

Nº 04  —  bioRxiv  —  Field report

Synthetic data recovers disease signatures in small studies

Model-generated samples recover prognostic disease signatures from transcriptomic datasets too underpowered to show them, according to a bioRxiv preprint. The approach, called generative augmentation, pads a small dataset with synthetic samples produced by a generative model.

Read →

 

· · ·

Reply with your discoveries. A human reads them. Forward freely.

Published by:

ARC