Method and system for normalization of gene names in medical text

Patent metadata

Field	Value
Publication number	US-2021319854-A1
Application number	US-201917272598-A
Country	US
Kind code	A1
Filing date	Aug 13, 2019
Priority date	Aug 28, 2018
Publication date	Oct 14, 2021
Grant date	—

Abstract

A method (100) for standardizing gene nomenclature, comprising: (i) receiving (110) a source; (ii) tokenizing (120) the source; (iii) comparing (130) a first token to a prefix tree structure with a root node, edges, and leaf nodes; (iv) determining (140) which edge extending from the root node to associated first leaf nodes the first token matches; (v) updating (150) an identification pointer with the location of the first leaf node; (vi) determining (160) which of one or more edges that a second token matches; (vii) updating (170) the identification pointer with the location of the second leaf node; (viii) repeating (172) the determining (160) and updating (170) steps with subsequent tokens until a subsequent token fails to match an edge extending from a leaf node or there is no edge extending from the leaf node; and (ix) providing (180) an identification of a canonical gene name.

First claim

1 . A computer implemented method for standardizing gene nomenclature, comprising:
receiving a source comprising one or more gene identifiers; tokenizing text from the source into a token stream; comparing a first token from the token stream to a data structure generated from a database of gene identifiers and corresponding canonical gene name for each of a plurality of genes, the data structure comprising a prefix tree structure with a root node, a plurality of edges, and a plurality of leaf nodes; determining which of one or more edges extending from the root node to associated first leaf nodes the first token matches; updating an identification pointer with the location of the first leaf node associated with the matching edge; determining which, if any, of one or more edges extending from the first leaf node to second leaf nodes that a second, subsequent token from the token steam matches; updating, if the second token matches an edge extending from the first leaf node, the identification pointer with the location of the second leaf node associated with the matching edge; repeating the determining and updating steps with subsequent tokens from the token stream using any additional extending edges and leaf nodes until a subsequent token fails to match an edge extending from a leaf node or there is no edge extending from the leaf node; and providing, when a subsequent token fails to match an edge extending from a leaf node, or if there is no edge extending from the leaf node, an identification of a canonical gene name based on a most recent location from the identification pointer.

2 . The method of claim 1 , further comprising:
generating a curated table of gene identifiers and associated canonical gene names; tokenizing the curated table into a token stream; and generating, using the token stream, the prefix tree structure.

3 . The method of claim 2 , wherein tokenization of the curated table into a token stream and the tokenization of text from the source into a token stream utilizes the same tokenization logic.

4 . The method of claim 2 , further comprising:
generating a list of most common words in a language; comparing the list of most common words to the gene identifiers in the curated table; and identifying any gene identifiers found in the list of most common words.

5 . The method of claim 1 , wherein providing an identification of a canonical gene name based on a most recent location from the identification pointer comprises:
updating the identification pointer with a location of a leaf node upstream of the most recent matching leaf node.

6 . The method of claim 1 , further comprising the step of determining whether the identified canonical gene name is also common natural language word.

7 . The method of claim 1 , wherein tokenization comprises generating an acronym for every capitalized phrase in a document, and preventing or removing each occurrence of a generated acronym from the token stream.

8 . The method of claim 1 , wherein tokenization comprises generating a single token of each string of Roman letters, generating a single token of each run of numbers, and generating a single token with each Greek letter.

9 . The method of claim 1 , wherein providing an identification of a canonical gene name comprises information about a location of a gene identifier in the source.

10 . A system for standardizing gene nomenclature, comprising:
a source comprising one or more gene identifiers; a data structure generated from a database of gene identifiers and corresponding canonical gene name for each of a plurality of genes, the data structure comprising a prefix tree structure with a root node, a plurality of edges, and a plurality of leaf nodes; and a processor configured to: (i) tokenize text from the source into a token stream; (ii) compare a first token from the token stream to the data structure; (iii) determine which of one or more edges extending from the root node to associated first leaf nodes the first token matches; (iv) update an identification pointer with the location of the first leaf node associated with the matching edge; (v) determine which, if any, of one or more edges extending from the first leaf node to second leaf nodes that a second, subsequent token from the token steam matches; (vi) update, if the second token matches an edge extending from the first leaf node, the identification pointer with the location of the second leaf node associated with the matching edge; (vii) repeat the determining and updating with subsequent tokens from the token stream using any additional extending edges and leaf nodes until a subsequent token fails to match an edge extending from a leaf node or there is no edge extending from the leaf node; and (viii) provide, when a subsequent token fails to match an edge extending from a leaf node, or if there is no edge extending from the leaf node, an identification of a canonical gene name based on a most recent location from the identification pointer.

11 . The system of claim 10 , wherein the processor is further configured to: generate a curated table of gene identifiers and associated canonical gene names; tokenize the curated table into a token stream; and generate, using the token stream, the prefix tree structure.

12 . The system of claim 10 , wherein the processor is further configured to: generate a list of most common words in a language; compare the list of most common words to the gene identifiers in the curated table; and identify any gene identifiers found in the list of most common words.

13 . The system of claim 10 , wherein the processor is further configured to update the identification pointer with a location of a leaf node upstream of the most recent matching leaf node.

14 . The system of claim 10 , wherein tokenization comprises generating an acronym for every capitalized phrase in a document, and preventing or removing each occurrence of a generated acronym from the token stream.

15 . The system of claim 10 , wherein providing an identification of a canonical gene name comprises information about a location of a gene identifier in the source.

Assignees

• Koninklijke Philips Nv

Inventors

• Pilato Samuel Frank
• Mankovich Alexander Ryan
• Mao Yong
• Yee Woei-Jye
• Wu Qingxin
• Van Ommering Robbert Christiaan
• Qin Weijian

Classifications

• G16B50/10 Primary

  Ontologies; Annotations · CPC title

• G06F40/166

  Editing, e.g. inserting or deleting · CPC title

• G06F40/284

  Lexical analysis, e.g. tokenisation or collocates · CPC title