By Allison Proffitt 

June 23, 2026 | When Eleanor Howe took the stage at Bio-IT World Conference & Expo 2026 to deliver the annual “Trends from the Trenches” address, she arrived with a clear mandate: no buzzwords, no vendor spin, and no marketing varnish.  

The session, a fixture of the conference since its founding more than two decades ago, was created by consultant Chris Dagdigian as a deliberate corrective to what he called “thinly disguised sales pitches delivered by a C-suite suit.” Since then, Dagdigian and others through the years have held firm to the candid and independent spirit of the session. This year, Dagdigian appeared via a pre-recorded message to formally pass the torch and introduced Howe as someone “cut from the same cloth”: honest, objective, and results-driven. 

Howe is no stranger to Bio-IT World, of course. A computational biologist and founder of Diamond Age Data Science, she has been part of the community for years herself: a regular speaker, awards program judge, and, most recently, host of the Trends from the Trenches podcast.  

She built this year’s Trends from the Trenches session from hours of interviews with colleagues, clients, and trusted advisors across the life sciences sector, delivering a wide-ranging, candid survey of where the field actually stands and what she predicts lies ahead.  

The One Thing Everyone Is Talking About: Coding Agents 

If there was a single trend Howe heard from every person she interviewed, it was the rise of AI-powered coding agents and their profound effect on the day-to-day work of bioinformaticians. 

“Bioinformaticians are able to write more code in less time, and they’re able to build bigger things in the same amount of time,” she said. “Biologists are able to write their own plots. They’re able to do some analysis—for better or for worse.”  

For worse: engaging with agents can be taxing, Howe warned. “Coding can be relaxing, but working with an agent is not.” But there are advantages as well. Howe pointed to Lior Pachter’s recent porting of edgeR into Python in a single week for $500 — a task that would otherwise have taken months (see his blog reflecting on the process). “He could carve out a week to do this work,” she said. “He could not carve out three months.” For a community that has long struggled to sustain open-source software on thin resources, she argued, this could be transformative. 

Where Does That Leave Scientists and Organizations? 

For those seeking to enter the field, though, Howe was blunt: the traditional on-ramp for talented coders entering the life sciences is closing. “Being somebody who just codes is not helpful in our field anymore. You need to know some science.” For junior job-seekers, the picture is even harder. “I really feel bad for the junior folks who are trying to get jobs in our field,” she said. “It is a tough, tough, tough job market out there.” 

Her advice to early-career scientists was pragmatic. Seek real-world experience. Learn to manage yourself, handle complex projects with interacting priorities, and persevere when everything breaks. “PhD programs [are] actually not bad at this,” she said. “The right thesis-based master’s program could do this as well.”  

Howe was equally direct with organizational leaders in the room. If your company has banned the use of chatbots and AI tools outright, she warned, your employees are already using the free versions and leaking intellectual property. “It’s really important to go and license something that’s going to protect your company,” she said, adding that organizations should resist the temptation to lock in with a single vendor. “I would say try to make [your platforms] LLM-agnostic and build a flexible interface that could be switch between many of them… Next year it’ll be totally different which one is the best for your current use case.” 

The “AI Drug Discovery” Problem: More Hype Than Substance 

Artificial intelligence is not somewhere out there autonomously discovering drugs and shepherding them through clinical trials, Howe assured the audience. What is actually happening, she argued, is more incremental and more interesting: a suite of distinct computational tools—lumped together under the catch-all label of “AI”—being integrated at various points along the drug discovery funnel. The structural biology work enabled by AlphaFold, she said, is a genuine and significant advance.  

She described one company that designed a novel modified antibody entirely in silico, grew it up in cells, obtained a crystallized structure, and found it matched the computational design within an angstrom. “That cuts a year, maybe two years off of development time,” she said. “It’s huge.” 

But Howe was equally pointed about the failures. She singled out transcriptional profiling models as a cautionary tale. The core problem is data: while AlphaFold succeeded because it could draw on a decades-old, rigorously structured protein database, transcriptional profiling involves tens of thousands of genes, each influenced by tissue type, age, sex, and environmental conditions too numerous to enumerate.  

What does work, she argued, are small, purpose-built models designed for specific biological questions with carefully curated, purpose-generated data. “The kitchen-sink method of collecting all of the data and putting it into a model and trying to build a foundation model out of that — it’s not going to work in biology the way it has worked for human language,” she said. “Biology is more complicated than English.”  

Some in the space have proposed synthetic data as a solution. Howe dismissed that idea with particular force. She warned that models trained on synthetic data learn the imperfections of the simulation, and training subsequent generations on that output compounds those errors exponentially. “Don’t simulate and try to train on that,” she said. “It’s a terrible idea. Make more or share with others.” 

A Warning on Bad Actors 

Yet sharing data is not without risk. She described hearing, firsthand, of companies that scrubbed attribution sections from scientific papers, presenting others’ work as their own. She also warned of organizations that approach potential collaborators requesting full access to proprietary data under assurances of good faith — assurances she said she was not confident all parties honor. 

Her practical recommendation: where possible, share model weights rather than raw data, similar to the federated model presented in one of the conference plenary sessions. Once a model has been trained and its weights extracted, the underlying dataset cannot be reverse-engineered.  

Things That Are Working, and Things That Are Not 

Howe ran through a rapid survey of additional trends, drawing consistent distinctions between what practitioners are actually using and what remains aspirational. 

On the “working” side: AlphaFold and related structural biology tools, active learning frameworks that guide experimental design to maximize information gain, and federated learning approaches that allow organizations to pool model weights without exposing underlying proprietary data. Also notable: AI-assisted data entry for electronic laboratory notebooks, which she described as an effective solution to scientists’ longstanding resistance to ELN data curation. “LLMs are the solution to their own problem,” she said. 

On the “not yet there” side: autonomous AI scientists (“I have not yet seen any AI scientists in action, and I’ve been inside a lot of companies”), virtual cells, and protein affinity modeling via new AI methods. “The old school stuff like Schrödinger is still what people are using for that.” 

She also offered a straightforward empirical finding: in head-to-head testing, classic machine learning methods — random forests, LASSO regression, linear models — frequently outperform the latest transformer architectures on specific biological tasks at a fraction of the cost and compute. “We have saved our clients a lot of money by saying, why don’t we try a linear model for this and see if that does what you need it to.” 

She was true to her stated purpose. “The goal today is not to predict the future with buzzwords,” she had said at the outset. “It’s to have an honest conversation about what the next era of digital R&D actually requires.” 

Listen to Howe’s full presentation—with takes on the data funding crisis, ‘omics technologies, virtual cells and more—on this month’s episode of the Trends from the Trenches podcast, or watch her presentation on YouTube.