---
title: 'Software Daily Summary: What Happened on November 17, 2026'
headline: Drug-discovery benchmarks test model performance against experiments
description: A computational chemistry roundup found low hit rates in compound-prediction
  tests. Infinitus’s CEO discussed AI agents for pharma field teams.
url: https://thebench.bio/news/software/2026-11-17
topic: software
date: '2026-11-17'
covers:
- '2026-11-17'
date_published: '2026-11-17T12:03:30Z'
date_modified: '2026-11-17T12:03:30Z'
tags:
- Software
- Industry
sources: 3
publisher: TheBench.Bio
---

# Software Daily Summary: What Happened on November 17, 2026

**Drug-discovery benchmarks test model performance against experiments**

A computational chemistry roundup found low hit rates in compound-prediction tests. Infinitus’s CEO discussed AI agents for pharma field teams.

## Today in brief

In CACHE #4, just two of 1,688 compounds selected by 23 pipelines were confirmed as hits, a 0.06% hit rate. [1] Infinitus CEO Ankit Jain said the company’s Dispatch agent helps field teams log information and get ready for meetings while traveling between appointments. [2]

## Drug-discovery benchmarks show gaps between model claims and measured performance

In a comparison of eight machine-learning docking models, AlphaFold3 and Boltz-2 localized active sites well across 210 targets, but each had a pose-recall hit rate below 10% at a strict 2 Å cutoff. [1] In CACHE #4, compounds selected by 23 pipelines had a 0.06% hit rate among 1,688 experimentally tested compounds. Two hits were confirmed across multiple binding assays, and one validated hit was described as chemically novel. [1]

A review of single-cell foundation models for perturbation modeling found that systematic benchmarks often showed no advantage over simple baselines, despite stronger results reported in the original studies. [1] Another study used SILCS to generate about 15,000 free-energy permeation profiles for 2,500 ligands across three model lipid bilayers. A separate analysis built a network of bioisosteric substituents from roughly 82,000 medicinal-chemistry series in ChEMBL. [1]

## Infinitus CEO describes AI agents supporting pharma field workflows

Infinitus CEO Ankit Jain said sales and field teams use the company’s Dispatch agent to enter information into customer relationship management systems and prepare for meetings while traveling between appointments. [2] Jain said the tool records data and follow-up tasks. He argued that AI agents could provide doctors’ offices with information beyond the hours when field teams are available. [2]

Jain described agents taking on roles similar to territory sales managers, medical science liaisons, and field reimbursement managers. He said pharma companies are considering how to serve more lower-tier provider networks. [2] [3] Jain said these tools should change more than productivity, including by giving field teams and physicians more ways to communicate. [2] [3]

## The day in numbers

- **0.06%**: Hit rate among 1,688 compounds predicted by 23 pipelines and tested in CACHE #4. [1]
- **Below 10%**: Pose-recall hit rates for AlphaFold3 and Boltz-2 at a strict 2 Å cutoff in an evaluation spanning 210 targets. [1]

## Sources

1. [This Week In Cheminformatics: Issue #047](<https://thisweekincheminformatics.substack.com/p/forty-seven>), This Week in Cheminformatics, 2026-11-17
2. [How Pharma Workflows Are Impacted by Agentic Models Beyond Productivity Gains](<https://www.pharmexec.com/view/workflows-impacted-agentic-models-beyond-productivity-gains>), Pharmaceutical Executive, 2026-11-17
3. [Agentic AI Models and Rethinking Pharma Field Teams](<https://www.pharmexec.com/view/agentic-ai-models-rethinking-field-teams>), Pharmaceutical Executive, 2026-11-16

---

TheBench.Bio wrote this from the reporting cited above, with a language model drafting and editing.

[1]: <https://thisweekincheminformatics.substack.com/p/forty-seven>
[2]: <https://www.pharmexec.com/view/workflows-impacted-agentic-models-beyond-productivity-gains>
[3]: <https://www.pharmexec.com/view/agentic-ai-models-rethinking-field-teams>
