iiinova — where your digital self keeps evolving

14/ In practice, extract passages from documents, align structured fields (gene names, lab values, etc.) with `taln`, and flag extractions that don't align. `taln` recovers alignments that other tools miss, reducing the number of extractions that need manual review.

12/ To summarize, verifying LLM-extracted text doesn't require another LLM. Subword tokenization + ordered alignment recovers ~50% more phrases than word-level tokenization, deterministically, in milliseconds. We release `taln`, BOAT, and BIO-BOAT as open-source tools and benchmarks.

11/ Unlike LCS and difflib, `taln` enumerates all valid alignments. On BOAT, that's ~8 and on BIO-BOAT ~90 (likely because biomedical terms like "VEGF-A165b" get split into 5+ subword tokens, combinatorially increasing matches.) Domain-specific tokenizers would help here.

100

10/ The remaining errors mostly come from subword boundaries that don't match how the phrase was originally selected. Edge tokens often have inconsistent start/end positions that change where the tokenizer splits the phrase.

100

9/ For non-contiguous phrases, we ablated one interior token from each target and reran alignment. Naive matching failed almost entirely (<1%). Word-level tokenization recovered ~50%. Subword tokenization recovered >96%.

8/ We tested word-level and subword tokenization with four alignment tools (taln, LCS, difflib, and naive matching.) On contiguous phrases, naive succeeds as expected, but word-level tokenization yielded ~50% missed alignments. Subword tokenization brought accuracy back to ~99%.

100

7/ With ordered alignment selected, we asked, how much does tokenization matter? We built the BOAT dataset (Berkeley Ordered Alignment of Text) from the Stanford Question Answering Dataset (SQuAD, by Pranav Rajpurkar and Percy Liang), containing 35K source-target pairs (and BIO-BOAT from biorxiv).

100

GitHub - sbooeshaghi/taln: non-contiguous token alignment

6/ Ordered alignment sits at the practical sweet spot in this hierarchy. It allows gaps while preserving order, keeping the number of alignments manageable. We implemented this in a Python tool called `taln`, inspired by pseudoalignment tools like kallisto. github.com/sbooeshaghi/taln

github.com

non-contiguous token alignment. Contribute to sbooeshaghi/taln development by creating an account on GitHub.

100

5/ Alignment methods form a hierarchy based on the constraints they impose on this map. Contiguous alignment -> ordered -> permutation -> rearrangement. Each relaxation admits more alignments but the search space grows fast—from linear to factorial to exponential.

100

4/ There's a long history of alignment methods in genomics and NLP. Pseudoalignment (@lpachter, @pmelsted uses set inclusion + k-mers. LCS uses order-preserving subsequences. To select an alignment method, we formalize alignment as a map from target to source.

3/ To find these phrases, you need an alignment method that allows gaps, and a tokenization strategy that doesn't merge the text you're looking for with surrounding punctuation (both matter!) We studied how much.

100

2/ Take this sentence from a scRNA-seq paper (by @ADHildreth.) An LLM correctly extracts "natural killer cells" and "CD96" as a cell-type marker gene pair. Both exist in the sentence, but the parenthetical "(NK)" breaks contiguity, so naive string matching fails.

100

Lior Pachter @lpachter.bsky.social · 11/02/2026

1/ LLMs are great at text extraction, but sometimes they hallucinate. A simple way to catch hallucinations is to check if the extracted text actually exists in the source. Turns out this is harder than it sounds. (new paper with Aaron Streets) www.biorxiv.org/content/10.6...

biorxiv.org

112

Reposted by sina b

Interesting article on LLM text extraction and the connection of that problem to (computational biology) sequence alignment. www.biorxiv.org/content/10.6... by @sina.bio and Aaron Streets.

biorxiv.org

081

sina b @sina.bio · 04/02/2026

I agree that machine readability enables these analyses in principle. Developing robust, grounded benchmarks will be essential for evaluating the utility of these kinds of analysis.

000

Reposted by sina b

Richard Sever @richardsever.bsky.social · 03/02/2026

"publication systems [should] distinguish between dissemination of results & communication of ideas, and optimize them separately. Results should be in explicit, machine-readable form, while narrative text serves as an interpretive layer for human readers" www.biorxiv.org/content/10.6...

biorxiv.org

1156

sina b @sina.bio · 03/02/2026

If you work in AI for Science, take a moment to familiarize yourself with a common failure mode: paranormal citations (or paracites). Our paper describes them.

082

Reposted by sina b

Lior Pachter @lpachter.bsky.social · 03/02/2026

AI hallucinations in science manuscripts are a nuisance. Paranormal citations, or paracites, will be a nightmare. www.biorxiv.org/content/10.6... (w/ @sina.bio & @lauraluebbert.com).

23211

sina b @sina.bio · 03/02/2026

Peer review is often opaque and confusing. @elife.bsky.social worked to change that. In a new preprint, we show how eLife’s Publish, Review, Curate model makes it possible to evaluate AI-generated reviews (with OpenEval) against human peer review. w/ @lauraluebbert.com and @lpachter.bsky.social

2189

Reposted by sina b

Gennady Gorin @goringennady.bsky.social · 06/03/2025

single-cell is a fun field. for instance, one of the heavily curated bixbench scenarios is about interpreting the results of a sc analysis and comparing to ground truth. this ground truth is, of course, based on DE analyses with some truly remarkable p-values for n=5

221

Reposted by sina b

C. Brandon Ogbunu @cbo.bsky.social · 06/03/2025

"No one is coming out of the sky to give you your grant money. Your citation portfolio won’t survive this market crash. Your credentials mean nothing. Everything is going to change." New for @undark.org undark.org/2025/03/06/o...

undark.org

How Science Can Adapt to a New Normal

Opinion | In the wake of attacks on the research enterprise, scientists need to focus on protecting its fragile infrastructure.

516879

sina b @sina.bio · 12/02/2025

“blocking retro nasal sensation with a nose clip significantly reduces the subjective and objective neural responses to sucrose taste”

000

Reposted by sina b

Richard McElreath 🐈‍⬛ @rmcelreath.bsky.social · 10/02/2025

Amazing to me how useful looking at data in 2D PCA continues to be, even though the approach sounds crazy on paper—"p-dimensional ellipsoid", rantings of a madman. PCA is the cockroach of dimension reduction. I expect it to be present in any advanced galactic civilization.

PCA can be thought of as fitting a p-dimensional ellipsoid to the data, where each axis of the ellipsoid represents a principal component. If some axis of the ellipsoid is small, then the variance along that axis is also small.

Figure 1b from https://www.nature.com/articles/s41586-024-08531-5/figures/1

38812

Reposted by sina b

C. Brandon Ogbunu @cbo.bsky.social · 06/02/2025

"One thing is certain: The changes we make ourselves will be healthier than the ones our adversaries demand." New work for @undark.org: undark.org/2025/02/06/o...

undark.org

The End of Science’s Peacetime

Opinion | Defending the practice of science from its adversaries will require dealing with some uncomfortable truths.

613278

Lastly, the fact that this announcement, from a historically legitimate scientific agency, was made exclusively on Twitter, on a Friday evening, is frankly, shameful.

020

What are the implications of this rate cut? I'm no expert but these outcomes may be impacted: - fewer faculty hires - department closures - fewer students (undergrad/grad), harder for int'l students - loss of US scientific hegemony

In FY2022, Caltech received $342,234,517 in federal funding (representing 80% of contract grant funding) 19.7% of which went to the NIH. A cut to 15% indirect costs, from 70%, means a loss of ~ $37.1M annually from NIH grants alone. Across all grants, $188M poof, gone.

100

As can be seen in the graphic, indirect costs vary per institution but previously were around 60-70%. This means for every dollar a professor was awarded from the NIH, the university got 60-70 cents. My prior institution, @caltech.edu had a 70% indirect cost rate.

100

When a professor at a university receives a research grant, they don't just get money for experiments and equipment. A percentage of the grant goes to the university as "indirect costs" - money used to "keep the lights on", administrative overhead, facilities, and infra.

100

sina b @sina.bio · 06/02/2025

🧵 On a Friday night, the NIH twitter account announced the most significant change to research funding in decades. What are indirect costs, how are universities funded and what are the impacts?

130

Is cause of “heavy traffic” known? Has this happened before? The thought of losing these resources is terrifying,.

sina b @sina.bio · 23/01/2025

Why are machine learning networks described as "neural"? The nomenclature arises because of the visual similarity between neural network architectures and connected neurons in the mammalian brain.

sina b @sina.bio · 23/01/2025

Reminder: there is little evidence that brain computation works in the same way as neural networks. Quote from "Understanding Deep Learning by Simon Prince (@simonprinceai.bsky.social)"

140

sina b @sina.bio · 23/01/2025

why is this desirable? Language models are helpful in parsing unstructured data, but QC reports are already structured...

020

sina b @sina.bio · 17/12/2024

Important to clearly define "genomic data"- certain data types may get better performance than others. Metadata storage seems to work well for my use-case. cc @manzt.sh

120

sina b @sina.bio · 17/12/2024

TIL about the watch command in the terminal: it reruns a command at set intervals, and is perfect for monitoring GPU usage or tracking real-time system updates.

EMBL-EBI @ebi.embl.org · 17/12/2024

Reposted by sina b

In this year's user survey, 89% of respondents said that EMBL-EBI data resources empowered them to undertake work that would otherwise not have been possible 💪 Big thank you to everyone who filled in the survey - we appreciate your input! Explore further findings: www.ebi.ac.uk/about/news/a...

ebi.ac.uk

Fuelling discovery together: 2024 user survey learnings

Blog post by Eleni Tzampatzopoulou, EMBL-EBI Impact Manager In summer 2024, EMBL-EBI ran a user survey, inviting our community to let us know how they use the open data resources we jointly manage wit...

0153

Reposted by sina b

Jase Gehring @skyjase.bsky.social · 08/12/2024

gonna post up in a cafe and speedrun a new protein diffusion model "from scratch" with Claude live poasting my way thru it, public Git repo last did this in June and it's really at the edge of both of our capabilities

041

Reposted by sina b

Jase Gehring @skyjase.bsky.social · 06/12/2024

I often make microfluidic emulsions and microparticles for scalable biology experiments, but the other day I was thinking, what if we polymerize the continuous phase of an emulsion? come learn about a delightful little branch of science, polyHIPEs!

121

sina b @sina.bio · 27/11/2024

I am honestly stoked by the ability to customize one's algorithm on @bsky.app. It's a killer feature (over Twitter) that makes it actually useable for scientific communication. Simply pick the feed you want to follow and boom you get the relevant content.