User:Etchevers/Notebook/Conference notes/2008/12/02

{| width="800"
 * style="background-color: #EEE"|[[Image:C14.jpg|128px]] Conference and seminar notes
 * style="background-color: #F2F2F2" align="center"|  |Main project page
 * style="background-color: #F2F2F2" align="center"|  |Main project page


 * colspan="2"|
 * colspan="2"|

Chris Ottolenghi
now H.U. (passed HDR) en biochimie métabolique et INSERM UMR-S747

New projet ANR will be submitted

Adduct formation.

Folate supplements prevent 50-60% NTDs. Causes of sensitivity: 10-30% MTHFR polymorphisms, maternal auto-antibodies to folate receptor (?) and other genes in mice.

What does “iatrogenic” mean?

Rescue of genetic defects by some metabolic pathways – up to six possible independent pathways. Pax3 rescue by folate or thymidine supplements, as Tead2 (folate); Axial defects mutant sensitive to methionine; Grhl3 (curly tail) inositol; Exencephaly on background SELH/Bc sensitive to certain diets (vitamin A?)

“One-carbon metabolic” cycle – ties cysteine, methionine, folates together. Betaine (converted by BHMT to DMG). The methionine part of cycle affect DNA methylation via effect on methyltransferases. Mthfr mouse mutants only limited NT effect in mice – slight curvature problem in lumbosacral tail only.

Homocysteine induces oxidative and ER stress in vitro, up in amniotic fluid if NTD. Adduction of homocysteine onto important proteins. Eg. thiolactone adduct which then can link to lysines and cause other problems by sticking onto other proteins (and cross-linking), or hemoglobin adduct. Retards fibrinolysis for example.

Adducts can induce an immune response as well. Eg. homocysteine is the antigen on the folate receptor that may

Try to make biotinylated reactive metabolites as probes and then streptavidin-based purification of proteins. Looking to find *all* proteins that might be targeted by adduction of homocysteine. Also using ion mobility mass spec to discriminate shape and size to augment chromography.

Use a panel of NTD fetuses and mothers, as well as mouse models.

I’m a little confused about technical approach. “Adductome” on amniotic fluid or maternal plasma using AFP or albumin; then also on supernatant of drug-treated hepatoma cell lines from mice. (Why –

Transcriptome approach on cell lines of what – amniotic fibroblasts? (A Budapest group has done some spina bifida versus not amniocytes… microarray analyses… but they can be comparable to other approaches.) What from the mother? And on the hepatomas which express AFP very highly. Endodermal derivatives also. But metabolic pathways are pretty universal.

Compared Rh NCC from chicken treated with homocysteine for 6h to the Budapest results? Looking for new “biomarkers” that may be more specific and sensitive than dosing AFP. Cf. Rick Finnell’s work http://www.ncbi.nlm.nih.gov/pubmed/17326132

Stan asks if the adducts are infinite? Chris says it is – non-enzymatic post-translational covalent modifications.

Possible to identify a protein via its unique spectrum and mass and size. Will find all that is abnormal, then find the sequence of the peptide, and eventually if there are adducts.

Want to establish a kind of library to cross with more biological animal models, as in another other –omics study.

Anne-Sophie
was at a conference on stats (Soc. Française de Statistiques)

1) Application of fashionable hidden Markov chains.

Observe sequence but the hidden fact is part of a gene, part of a promoter, etc. Information used is probability of transition between one and another nucleotide, can include modules or use exceptionality scores eg. a motif in a sequence like a binding site. So method can help find this kind of motif.

Alignments – maximizing scores with a certain number of algorithms eg. BLAST, FASTA, CLUSTAL. Now can use a more statistical approach – the Markov chain is 3 states – match, mismatch or indel. Sample of possible alignments that are possible between species, not the reconstruction of whole genome sequencing.

2) Multiple testing corrections.

Correction of Bonferroni – to not have too many false-positives, apply correction when have multiple tests. But perhaps too conservative. Not powerful tests even if there is a true significance. Also doesn’t take into account interdependence of tests.

If m tests with level = 5%, false positives if 10 independent tests will be 50%. Benjamini and Hochberg to control how many false positives.

FDR (false discovery rate) are more powerful tests of how to predict real incidence of false positives.

3) Local score method. If association – multiple signals at same place because of linkage disequilibrium. Make a score by a region (a locus) rather than each marker independently. Algorithm is called LHiSA and it is really helpful for replications especially between one population and another.

CGH – noise and non-linear fluorescence, mosaicism in sample. Therefore tendency to underestimate CNVs.

How segment genome so to better make this estimation? Using hidden Markov chain again – how many hidden states? 1 copy, 2, 3, none, for a given point in the distribution of BAC intensity and position in genome? Another technique is to segment the genome starting with natural breakpoints. What is first variation within a segment of non-copy-variant? Since there are many combinations possible, adapt the model to reduce number of segments. This is more successful than the Markov chain approach.

LASSO approach for finding expression data relationships between gene transcripts. But for genes with lots of relationships not very powerful (think p53).

Models of networking – facing a complex graph, how find groups or hubs? Sociologists use a program called MIXNET.

Tania mentions Jean-Christophe Fournié (chef de service Anapath) – cf malformations.org site to try to classify congenital malformations with this same program.


 * Heather 08:27, 5 December 2008 (EST):


 * }