Applied text mining
Download as PPT, PDF1 like405 views
This document discusses natural language processing and text mining techniques for biomedical literature and electronic health records. It describes named entity recognition to identify concepts like genes and proteins, relation extraction to find interactions between entities, and information extraction to formalize stated facts. It also discusses integrating extracted information with structured databases and visualizing relationships through web interfaces. Medical text mining can apply these techniques to clinical notes to identify diseases, drugs, adverse events and more for applications like comorbidity analysis, patient stratification, and pharmacovigilance.
![Gene and protein names
Cue words for entity
recognition
Verbs for relation extraction
[nxexpr The expression of
[nxgene the cytochrome
genes
[nxpg CYC1 and CYC7]]]
is controlled by
[nxpg HAP1]](https://image.slidesharecdn.com/jensen2014talk5-140409095004-phpapp02/85/Applied-text-mining-82-320.jpg)
Applied text mining
- 2. >10 km
- 3. too much to read
- 5. ~40 seconds per paper
- 6. computer
- 7. as smart as a dog
- 8. teach it specific tricks
- 17. find the relevant papers
- 18. ad hoc retrieval
- 20. “yeast AND cell cycle”
- 21. PubMed
- 23. indexing
- 24. fast lookup
- 25. stemming
- 26. word endings
- 28. MeSH terms
- 29. Mitotic cyclin (Clb2)-bound Cdc28 (Cdk1 homolog) directly phosphorylated Swe1 and this modification served as a priming step to promote subsequent Cdc5- dependent Swe1 hyperphosphorylation and degradation
- 30. no tool will find that
- 33. Mitotic cyclin (Clb2)-bound Cdc28 (Cdk1 homolog) directly phosphorylated Swe1 and this modification served as a priming step to promote subsequent Cdc5- dependent Swe1 hyperphosphorylation and degradation
- 35. CDC2
- 36. cyclin dependent kinase 1
- 40. CDC2
- 41. Cdc2
- 43. cyclin dependent kinase 1
- 45. name expansions
- 47. CDC2
- 48. hCDC2
- 49. “black list”
- 50. SDS
- 51. efficient tagger
- 52. Pafilis et al., PLOS ONE, 2013
- 53. benchmarking
- 54. the formal way
- 57. precision
- 58. recall
- 59. much work
- 61. random sampling
- 63. precision
- 64. no recall
- 65. much less work
- 67. Mitotic cyclin (Clb2)-bound Cdc28 (Cdk1 homolog) directly phosphorylated Swe1 and this modification served as a priming step to promote subsequent Cdc5- dependent Swe1 hyperphosphorylation and degradation
- 68. Mitotic cyclin (Clb2)-bound Cdc28 (Cdk1 homolog) directly phosphorylated Swe1 and this modification served as a priming step to promote subsequent Cdc5- dependent Swe1 hyperphosphorylation and degradation
- 69. Reflect
- 70. reflect.ws
- 73. Mitotic cyclin (Clb2)-bound Cdc28 (Cdk1 homolog) directly phosphorylated Swe1 and this modification served as a priming step to promote subsequent Cdc5- dependent Swe1 hyperphosphorylation and degradation
- 74. two approaches
- 75. the formal way
- 80. semantic tagging
- 81. sentence parsing
- 82. Gene and protein names Cue words for entity recognition Verbs for relation extraction [nxexpr The expression of [nxgene the cytochrome genes [nxpg CYC1 and CYC7]]] is controlled by [nxpg HAP1]
- 84. high precision
- 85. poor recall
- 89. co-mentioning
- 90. counting
- 91. within documents
- 93. within sentences
- 94. quality score
- 97. high recall
- 98. high precision
- 100. unknown type
- 102. STRING
- 103. protein associations
- 104. string-db.org
- 105. STITCH
- 106. STRING + 300k chemicals
- 107. stitch-db.org
- 108. COMPARTMENTS
- 111. TISSUES
- 112. tissue expression
- 114. DISEASES
- 117. curated knowledge
- 118. pathways
- 119. Letunic & Bork, Trends in Biochemical Sciences, 2008
- 120. experimental data
- 121. gene expression
- 124. gene neighborhood
- 125. Korbel et al., Nature Biotechnology, 2004
- 126. many databases
- 127. different formats
- 129. variable quality
- 130. not comparable
- 131. hard work
- 132. common identifiers
- 133. quality scores
- 134. score calibration
- 135. visualization
- 136. web interfaces
- 137. bulk download
- 138. why so many resources?
- 139. Swiss army knife syndrome
- 141. medical text mining
- 144. opt-out
- 145. opt-in
- 146. structured data
- 147. Jensen et al., Nature Reviews Genetics, 2012
- 148. unstructured data
- 149. clinical narrative
- 151. Danish
- 152. busy doctors
- 153. psychiatric patients
- 155. custom dictionaries
- 156. diseases
- 157. drugs
- 158. adverse events
- 159. expansion rules
- 160. phonetic spelling
- 161. typos
- 162. sentence filters
- 163. negations
- 164. family members
- 165. delutions
- 167. Roque et al., PLOS Computational Biology, 2011 3262638254947 Assigned codes Text mined codes
- 168. comorbidity
- 169. Roque et al., PLOS Computational Biology, 2011
- 171. Roque et al., PLOS Computational Biology, 2011
- 172. pharmacovigilance
- 175. Eriksson et al., submitted, 2013
- 176. EMBO Practical Course Computational Biology: Genomesto Systems Puerto Varas, 3-9April2014 Thank you!Thank you!