machine learning methods for analysis of the global fossil record
TRANSCRIPT
Machine learning methods foranalysis of the global fossil record
Lecture 1
Indrė Žliobaitė[email protected]
Photo credit: Kayle Reed
Biospheric dataFossil databases
Environment observation sensorsSatellite imaging data
Ecological and conservation data
Weather stations
https://www.atm.helsinki.fi/SMEAR/index.php/smear-ii
http://www.iucnredlist.org/
http://www.worldclim.org/methods1
https://neo.sci.gsfc.nasa.gov/view.php?datasetId=MCD12C1_T1
What can be answered using machine learning
● Reconstructing how life was in the past and how life works in general
● Analyzing relations between organisms and environments,understanding biospheric change processes
● Analyzing why and how species evolve
● Understanding circumstances of early human evolution
Environment:TemperaturePrecipitationProductivitySeasonalityVegetation typeWoody cover...
Occurring animals:Species ASpecies BSpecies C...
Features of animals:Body massSlim legs?Sharp teeth?...
Why understanding the past?
Source: https://climate.nasa.gov/evidence/
Climate change
Source: https://climate.nasa.gov/evidence/
Extreme events
Reconstructing and analyzing climate over time
Fortelius et al 2002
Source: http://humanorigins.si.edu/evidence/human-family-tree
Circumstances of early human environments
Circumstances of early human environments
Image credits:Mauricio Anton
Source: https://www.nature.com/nature/journal/v507/n7492/full/507303a.html
Slide credit: Mikael Fortelius
Estimates for the past, the Turkana basin
Fortelius et al 2016
Ecometric modeling makes quantitative analysis possible,complements qualitative insights with quantitative reasoning
Paleobiology and conservation:putting dead to work
● “The geologic record as anatural ecological andevolutionary laboratory”
Dietl and Flessa 2011
Source: Barnosky et al 2017
Source: Turkana Basin Institute
Seminar agenda and topics
To pass the seminar
● Short presentation of a research article
● Data analysis project + presentation
– Implement one method from the article and one modification
● Attendance
● Exam?
Organization
● Mondays: lectures or presentations
● Tuesdays: project consultations
– come to discuss your projects, for help in interpreting articles,accessing data, implementation and analysis
A fully funded PhD position is available at the University of Helsinki on computational methods for analysis of the global fossil record, ecological and climate data to better understand history of life, evolutionary processes and environmental change.
Apply via UH by October 31, 2017https://www.helsinki.fi/en/open-positions/doctoral-student-machine-learning-for-evolving-data
Master projects are welcome(contact me)
+
Topics for machine learning
1) Chronofaunas
2) Ecometrics
3) Macroevolution
4) Phylogenetic regression
5) Phylogenetic trees
Chronofaunas:(mostly unsupervised learning)
Eronen et al 2009
tracking faunal communities over time
Chronofaunas:(mostly unsupervised learning)
identifying chronofaunas
Bingahm and Mannila 2014
Chronofaunas:(mostly unsupervised learning)
seriation - ranking of localities in time
Fortelius et al 2006 Puolamaki et al 2006
World today
Environment:TemperaturePrecipitationProductivitySeasonalityVegetation typeWoody cover...
Occurring animals:Species ASpecies BSpecies C...
Features of animals:Body massSlim legs?Sharp teeth?...
Fossil sites1-5 million years old
Fossil sites20-30 million years old
......
......
Occurring animals:Species XSpecies YSpecies Z...
Features of animals:Body massSlim legs?Sharp teeth?...
one observation/ learning instance
Ecometrics - analyzingecosystems of the present and the past
Estimates for the past, the Turkana basin
Fortelius et al 2016
Age in MA Age in MA
Why animal traits predict climate?
http://www.vivo.colostate.edu/hbooks/pathphys/digestion/pregastric/horsepage.html
http://home.utah.edu/~u0230781/project/foresta/pages/moose_jpg.htm
https://www.etsy.com/au/listing/163309116/2-huge-genuine-moose-teeth-molars-alces
Grazing in open environments requires much more more tooth than browsing in a closed forest
Features of animal teeth
Žliobaitė et al 2016
● Hypsodonty (HYP)
● Horizodonty (HOD)
● Presence of acute lophs (AL)
● Presence of obtuse lophs (OL)
● Structural fortification of cusps (SF)
● Occlusal topography (OT)
● Coronal cementum (CM)
Predicting climate from animal communities(concept drift, transfer learning)
Satellite observation data Model fit
Ecometrics:(mostly predictive modeling, transfer learning)
Liu et al 2012
predicting productivity of environment from animal teeth
Fortelius et al 2016
Global regression, main traits
Advanced traits, focus on Africa
Ecometrics:(mostly predictive modeling, transfer learning)
Redescription minning –local relations between climate and dental traits
Galbrun et al in review
Galbrun et al in review
Macroevolution:(probabilistic modeling, maximum likelihood)
Analyzing tempo and mode of evolution
Voje 2016
Macroevolution:(probabilistic modeling, maximum likelihood)
Analyzing tempo and mode of evolution
Raia et al 2016
Macroevolution:(probabilistic modeling, maximum likelihood)
Estimating probabilities of occupancy
Foote 2016
Phylogenetic regression
Freckleton et al 2002 Felsenstein 1985
2017
Phylogeny - reconstructing the tree of life
2017
Fitting phylogenetic trees
Baron et al 2017
Fitting phylogenetic trees
Baron et al 2017
Fossils and fossil databases
https://fineartamerica.com/featured/1846-richard-owen-and-moa-leg-fossil-paul-d-stewart.html
Richard Owen and MOA leg fossil, 1846
Wikimedia commons
Museums – catalogues – books – fossil databases
Recorded in global fossil databases, Including: taxonomic identification, fossil age, location, other characteristics
Imgage: http://www.helsinki.fi/science/now/ From Heck's Iconographic Encyclopedia (1851),http://users.dickinson.edu/~nicholsa/Romnat/fossils.htm
Sepkoski 1981
Fossils
Fossils are collected all over the world- depositional environment needs to be specialfor fossil preservation- dating / stratigraphy
Photo credit: Mikael Fortelius
Photo credit: Mikael Fortelius
Fossil data
Image: http://africanfossils.org/search
Described and catalogued- identification of species- measured traits- environmental context- 3D scans
NOW database (Helsinki)
http://www.helsinki.fi/science/now/
PDBD – Paleobiological database
https://paleobiodb.org
Slide credit: Laura Säilä
Slide credit: Laura Säilä
Fossil teeth
Slide credit: Laura Säilä
NOW database (Helsinki)
http://www.helsinki.fi/science/now/