I happen to read this thread 3 years ago and got interested in Prof. Jim's statement that " It was also asked about computing heritability “for each landmark”. That would be difficult without some strong assumptions."
I understand there has been discussion in this forum about the inappropriateness of analysis of landmarks one at a time after GPA. However, I am curious to learn under what "strong assumption" could we perform analysis "one landmark at a time"? Shape by definition is the configuration of ALL landmarks after GPA. This means one should never analyze landmarks one at a time with the slightest hope that Procrustes shape coordinates for a single landmark represent local shape information. I think understanding the exceptional condition where one could analyze landmarks one at a time helps me gain deeper understanding of this issue. Thank you. Kieran On Wednesday, July 31, 2019 at 9:49:38 PM UTC+8 [email protected] wrote: > Dear Prof. Houle, > > Thank you very much for your interest in my questions. My dataset contains > 3D body shape data of people from peopole of two ethnicities. Based on such > phenotypic data, I wish to estimate population differentiation and > selection differential/gradient, and total predicted response to selection. > I agree with you that estimation of these quantities requires phenotypic > and additive genetic variance-covariance matrix rather than heritability. I > have only phenotypic data at the moment, although I will obtain the genetic > data sometime later. > > I understand that one approach to estimate the G matrix is to multiply the > P matrix by an estimated value of heritability. Implicit in this approach > is the assumption of proportionality. Am I right that it is impossible to > estimate G in the absence of genetic data, apart from the approach above? > > Regards, > Patrick > > On Wednesday, 31 July 2019 03:10:31 UTC+8, dhoule wrote: > >> First of all, quantitative genetic concepts are defined relative to a >> reference or base population. If you are mixing individuals from a wide >> geographic area, what is your population that you are making inferences >> about? >> >> Second, you can't do any quantitative genetics unless you have both >> 'unrelated' (relative to the defined reference population), and individuals >> with known degrees of relationship. Do you have genomic data on these >> individuals? >> >> If all you have are individuals of unknown relationship, all you can do >> is calculate the P matrix, then make some very strong assumption - like >> heritability is 0.3, and get a sort of proxy G matrix. This might be >> better than nothing. >> >> Why do you want to know the heritability? To predict the magnitude of >> changes, you need the genetic variance and covariances, rather than >> heritabilities. >> >> David Houle >> On 7/30/2019 11:59 AM, lv xiao wrote: >> > A huge thank you to all who have contributed to my questions. It did took >> me a while to read the articles kindly suggested. I now have some basic >> understanding of heritability, selection differential, selection gradient, >> etc. >> >> I have 3D data on body shape of Europeans and Africans and I wish to >> obtain estimates of heritability and visualize shape changes associated >> with selection differential and response to selection. The first thing I >> would like to do is to estimate the additive genetic variace-covariance >> matrix G, and the phenotypic variance-covariance matrix P. Software such as >> ACE and Wombat have been used for such purpose. The essence of this seems >> to be a multivariate mixed-effect model where the dependent variable is the >> shape data in matrix and the independent variables consist of both fixed >> and random effects. However, my dataset is not related to full-sibling >> design and I would like to ask how could I derive matrix P and G from my >> phenotypic data? >> >> Once these two matrices are estimated, I will be able to performe further >> analyses on my own. Thank you. >> >> On Wednesday, 24 July 2019 21:12:13 UTC+8, lv xiao wrote: >>> >>> Dear all, >>> >>> Heritability is traditionally estimated based on genetic data. However, >>> it has also been estimated from phenotypic data. In this case, twin >>> comparison is the common study design. My question is if I have a sample of >>> *indenpendent >>> individuals* whose 3D shape data are available, is it possible to >>> *calculate >>> heritability based on Procrustes shape coordinates* for overall shape >>> and for each landmark? >>> >>> Several studies have used the Sequential Oligogenic Linkage Analysis >>> Routines (SOLAR) package for estimation of heritability of craniofacial >>> trait based on independent samples (J. Anat.(2009) 214, pp19–35; THE >>> ANATOMICAL RECORD 298:1535–1547 (2015)). However, these traits are linear >>> distances, which are quite different from "shape" based on Procrustes shape >>> coordinates. >>> >>> Thanks >>> >> -- >> You received this message because you are subscribed to the Google Groups >> "Morphmet" group. >> >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> >> >> To view this discussion on the web visit >> https://groups.google.com/d/msgid/morphmet2/6d5d4ad1-936f-4a2a-a33b-3445a59a6373%40googlegroups.com >> >> <https://groups.google.com/d/msgid/morphmet2/6d5d4ad1-936f-4a2a-a33b-3445a59a6373%40googlegroups.com?utm_medium=email&utm_source=footer> >> . >> >> -- >> David Houle >> Department of Biological Science >> Florida State University >> 319 Stadium Drive >> Tallahassee, FL 32306-4295 >> 850-339-5670 <(850)%20339-5670>http://bio.fsu.edu/~dhoule/ >> >> -- You received this message because you are subscribed to the Google Groups "Morphmet" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/morphmet2/aab303e0-2433-41c1-9175-b3351fdb2c04n%40googlegroups.com.
