Get the best results from AI-powered semantic search across 19M+ PubMed articles.
Papedex uses a multi-channel search engine that combines two fundamentally different approaches to find relevant papers:
Your query is converted into a mathematical representation (embedding) and compared against pre-computed embeddings of every title and abstract in PubMed. This finds papers that are conceptually similar to your query, even if they use different words. For example, searching "heart attack" will also find papers about "myocardial infarction" and "acute coronary syndrome."
Simultaneously, your query is matched against titles and abstracts using traditional keyword matching (BM25 ranking). This ensures papers that use your exact terminology are always found.
If your query looks like an author name, Papedex searches the author index directly. See Searching for Authors for details.
Results from all channels are merged using Reciprocal Rank Fusion (RRF), which combines rankings from different sources into a single ranked list. A paper that ranks well across multiple channels (e.g., both semantically and by keywords) gets boosted to the top. Recent papers receive a gentle boost, reflecting that newer research is often more relevant.
| Instead of | Try | Why |
|---|---|---|
| diabetes | type 2 diabetes insulin resistance mechanisms | Gives the semantic model more context to understand what you actually want |
| cancer | pancreatic cancer immunotherapy checkpoint inhibitors | Narrows from 3M+ papers to your specific area |
| CRISPR | CRISPR-Cas9 gene editing therapeutic applications | Distinguishes from CRISPR review papers, bacterial biology, etc. |
Semantic search understands concepts, not just keywords. You can search in a more natural way:
| Query | What it finds |
|---|---|
| how does sleep affect memory consolidation | Papers on sleep-dependent memory processes, even if they don't use these exact words |
| proteins that cause Alzheimer's disease | Amyloid-beta, tau, presenilin papers without needing to know those terms |
| side effects of mRNA vaccines | Adverse events, reactogenicity, safety profiles of COVID-19 and other mRNA vaccines |
You can mix natural language with exact terminology and field-specific syntax:
| Query | Effect |
|---|---|
| gut microbiome depression journal:nature | Semantic search for concept + journal filter |
| author:doudna CRISPR delivery year:2022-2025 | Author + topic + date range |
| "gut-brain axis" AND inflammation | Exact phrase + boolean AND |
Prefix a term with a field name and colon to restrict the search to that field:
| Syntax | Description | Example |
|---|---|---|
| title:term | Search titles only | title:CRISPR |
| author:name | Search authors only | author:doudna |
| abstract:term | Search abstracts only | abstract:neuroplasticity |
| journal:name | Filter by journal | journal:nature |
| year:range | Filter by year or year range | year:2020-2025 or year:2024 |
Field prefixes can be combined with free text: author:hattar melanopsin circadian
Wrap a phrase in double quotes to find exact matches:
| Syntax | Effect |
|---|---|
| "gut-brain axis" | Finds papers with this exact phrase in title or abstract |
| "machine learning" radiology | Exact phrase + free keyword |
Use AND, OR, and NOT (case-insensitive) to combine terms:
| Operator | Example | Effect |
|---|---|---|
| AND | CRISPR AND cancer |
Results must match both terms (each searched independently, results intersected) |
| OR | Alzheimer OR dementia |
Results can match either term (results merged) |
| NOT | diabetes NOT type 1 |
Results for first term, excluding papers mentioning second term |
Boolean operators must be uppercase or will be treated as regular search terms. Multi-word boolean queries search each side as a full phrase.
Use the filter panel (click "Show Filters") or the year: syntax:
year:2024 — Papers from 2024 onlyyear:2020-2025 — Papers from 2020 through 2025Use the filter panel or journal: syntax. Partial matches work:
journal:nature — Nature, Nature Medicine, Nature Genetics, etc.journal:cell — Cell, Cancer Cell, Molecular Cell, etc.Use the slider to adjust how many results are returned (5 to 100). More results take slightly longer but cover a wider range.
Papedex and PubMed have different strengths. Understanding these helps you get the best results from each:
| Aspect | Papedex | PubMed |
|---|---|---|
| Search method | Semantic (AI embeddings) + keyword (BM25), fused | Keyword + MeSH controlled vocabulary + ML ranking |
| Short queries | Less effective — add context for better results | Strong — auto-expands using MeSH synonyms |
| Descriptive queries | Excellent — semantic model uses all context | Good, but may over-weight certain MeSH terms |
| Concept discovery | Strong — finds papers using different terminology for the same concept | Moderate — limited to MeSH term relationships |
| Author search | Name matching (no disambiguation for common names) | Disambiguated via ORCID and MeSH Author IDs |
| Recency | Gentle recency boost (~3%/year decay) | Strong recency signal in "Best Match" |
Begin with a detailed query describing what you're looking for, then use filters and field syntax to refine:
gut microbiome influence on depression and anxietyyear:2020-2025journal:natureWhen you need a specific phrase, quote it: "gut-brain axis" ensures those words appear together.
Instead of just Zhang F, use author:zhang CRISPR or Feng Zhang gene editing to find the right researcher.
Enable "Compare with PubMed" to see if important papers appear in one engine but not the other. This is especially useful when evaluating your search strategy.
For exploratory searches, increase the results slider to 50-100. For targeted searches, 10-25 is usually sufficient.