APTLAS: An Indexed APT Literature Repository

APTLAS is a specialized, indexed repository and web tool containing approximately 2,300 Atom Probe Tomography publications with domain-specific metadata, designed to overcome the limitations of general search engines by enabling precise filtering and discovery of APT literature based on material systems, instruments, and analytical approaches.

The Gist

Imagine you are a chef trying to find a specific recipe for a dish made with "boron-doped silicon" cooked on a "LEAP 5000 XS" stove. If you just type that into a general search engine like Google, you might get thousands of results, but most of them won't tell you exactly what kind of stove was used or the specific temperature settings. You'd have to read through hundreds of articles just to find the one that matches your exact needs.

This is the problem the authors of this paper are solving for scientists who use a technique called Atom Probe Tomography (APT). APT is a high-tech microscope method used to see materials at the atomic level. Over the last 20 years, the number of scientific papers about APT has exploded, scattered across hundreds of different magazines and journals. Finding a specific piece of research based on the exact machine used or the specific material tested has become like finding a needle in a haystack.

Here is a simple breakdown of what they built, APTLAS:

1. The Problem: A Messy Library

Think of the world of APT research as a massive, chaotic library where books are thrown on the floor. The books are about everything from metals to biological materials, and they were written using different machines. If you ask a librarian (a standard search engine), "Show me books about silicon," they might give you a list, but they won't tell you which books were written using a specific type of laser or a specific machine model. The "metadata" (the details about how the experiment was done) is lost in the general search.

2. The Solution: A Smart, Organized Index

The authors created APTLAS, which acts like a super-organized, digital card catalog for this specific library.

• What it is: A database containing about 2,300 published papers.
• How it works: Instead of just listing titles, they extracted specific details from every single paper, such as:
  • What material was studied? (e.g., Metals, Semiconductors, Rocks)
  • What machine was used? (e.g., LEAP 5000 XS)
  • How was it done? (e.g., Laser settings, temperature, pulse rate)

3. How They Built It: The Three-Step Assembly Line

The team built this database using a three-step process, similar to a factory assembly line:

1. Gathering: They used a computer script to ask a global database (CrossRef) for every paper mentioning "Atom Probe" since 2001.
2. Reading and Sorting (The AI Part): This is the clever part. They used a "Large Language Model" (an AI that reads text) to read every paper. They gave the AI a specific checklist (a schema) and asked it to pull out the details (like the machine name or laser type) and put them into a neat digital file. If a paper didn't mention a detail, the AI marked it as "unknown" rather than guessing.
3. Cleaning Up: They ran a final check to remove duplicate papers, throw out non-research documents (like error corrections), and fix any obvious mistakes.

4. The Tool: A User-Friendly Website

The result is a free, simple website (a "single-page app") that anyone can use.

• The Interface: Imagine a dashboard with five ways to start your search: by Type of paper, Material, Application, Instrument, or just a Keyword.
• The Filters: You can narrow down your search instantly. For example, you could filter for: "All papers about Semiconductors analyzed on a LEAP 5000 machine using a Laser."
• The Results: You get a list of cards. You can click a card to see the full details, the abstract, and a link to the original paper. You can also check boxes to select multiple papers and export them as a simple list.

5. What They Admit (The Limitations)

The authors are honest about the tool's current limits:

• AI isn't perfect: The AI that read the papers is very good, but not 100% error-free. Sometimes it might miss a specific number or get a detail wrong. They advise users to double-check the original paper if they need perfect accuracy.
• Categories aren't always clear: Some research fits into multiple categories (like a paper that is both about a new machine and a new material). The system has to force these into one box, which can sometimes feel a bit subjective.

The Bottom Line

APTLAS is a curated, searchable index designed to save scientists time. It takes a chaotic, rapidly growing field of research and organizes it into a tool where you can filter by the exact variables that matter most to your experiment. It doesn't do the science for you; it just helps you find the right "recipe" so you don't have to read the whole library to find it.

The tool is available online, and the database files are open for anyone to download and use on their own computers.

Technical Summary

Technical Summary of APTLAS: An Indexed APT Literature Repository

Problem Statement
Atom probe tomography (APT) has expanded rapidly over the past two decades, moving beyond its historical roots in metallurgy to encompass a diverse range of materials and applications. This growth has resulted in a literature dispersed across general-interest journals (e.g., Nature, Science), specialized microscopy publications, and numerous materials-specific journals. While conventional search engines like Google Scholar are effective for general retrieval, they fail to preserve or filter by domain-specific metadata critical to APT research. Specifically, standard search tools cannot effectively filter by experimental variables such as instrument configuration, laser wavelength, analysis mode, or specimen preparation methods. This limitation creates a significant impediment for researchers attempting to identify prior work based on specific combinations of variables, a task complicated by the high dimensionality of the APT parameter space and the strong dependence of measurement results on experimental conditions.

Methodology
The authors constructed APTLAS through a three-stage pipeline: literature acquisition, metadata extraction, and validation/assembly.

1. Literature Acquisition: A candidate list of publications was assembled by querying the CrossRef API using keywords including "Atom Probe Tomography," "Atom Probe Microscopy," "Field Ion Microscopy," "Laser-Assisted Atom Probe," and "Local Electrode Atom Probe." The query primarily targeted publications from 2001 onwards, though select foundational works predating this range were included.

2. Categorization Schema and Metadata Extraction: A hierarchical schema was designed to balance granularity with tractability. Publications were indexed across five categories:

   • Publication Type: Journal Article, Letter, Review, Conference Proceeding, Thesis.
   • Topic: Ranging from Instrumentation and Technique Development to Materials Science and Application Studies.
   • Material System: Including Metals, Semiconductors, Ceramics, Functional Materials, Thin Films, Nuclear Materials, Geological Materials, and Biomaterials.
   • Instrument: Name, configuration, and laser wavelength.
   • Analysis Conditions: Analysis mode, base temperature, pulse rate, detection rate, and pulse fraction/energy.

   Extraction was performed using a local Llama 3.1 model served via Ollama. Each publication was processed with a structured-output prompt to return a JSON object conforming to the schema. Fields were populated from the text or set to null if information was not reported.

3. Validation and Assembly: Extracted records underwent validation to remove duplicates (identified via content hashing and DOI matching) and non-research documents (e.g., errata, supplementary files). The final dataset, containing approximately 2,300 records, was assembled into a single data.json file.

Key Contributions and Results
The primary contribution is the deployment of APTLAS, a curated index of published APT literature accompanied by a web-based navigation tool.

• Database Scale: The current database contains ~2,300 records, each accompanied by extracted metadata.
• Web Tool Functionality: The tool, hosted at https://aptlas.bavleyguerguis.com/, is a single-page HTML application. It offers five entry points: browsing by Type, Material, Application, Instrument, and keyword search.
• Interactive Features: Users can filter results using a combination of secondary filters (material system, instrument, topic, journal, year). The interface allows for the selection of individual records via checkboxes, exporting selected records as a plain-text citation list, sorting results by relevance, year, or alphabetically, and flagging records with suspected extraction errors for review.
• Detailed Views: Clicking a record reveals a detailed view of all extracted metadata, including the abstract and a DOI hyperlink.

Significance and Claims
The paper positions APTLAS as a practical solution to reduce the friction associated with identifying prior work in a niche but rapidly growing technique. The authors claim that by preserving domain-specific metadata often lost in general search engines, the tool facilitates the identification of work based on specific experimental combinations (e.g., laser-pulsed APT of boron-doped silicon on a specific instrument model).

The authors maintain a modest stance regarding the system's current limitations. They acknowledge that the LLM-based extraction is not error-free, particularly concerning experimental parameters, and encourage users to verify metadata against original publications when accuracy is critical. They also note that the categorical schema is a necessary simplification of a complex literature, which may lead to subjective or erroneous assignments for publications that do not fit cleanly into a single category.

Future work is outlined as focusing on periodic re-querying of CrossRef for new publications, refining extraction methods with more capable models, and assessing more granular extraction targets, such as specific dopant species. The code and data (including index.html and data.json) are made available for local deployment, and the extraction pipeline scripts are available via the project repository.