Re:Wrong target market.
on
RFID Cookware
·
· Score: 1
You know, you can't get a stove that puts out BTUs like a professional chef's stove. Induction cooking allows you to achieve those kinds of heat outputs (as it only heats the cookware, not the stove, walls neat the stove, etc) in your own kitchen.
That's why I want induction cooking - I can't stir fry food in a wok nearly as well as I'd like with my wimpy consumer stove. As for RFID to control temperature - that might be really useful interested in holding food at a certain temperature to cook, like in vaccume cooking (sous vide). In that case, you don't cook food above 175 degrees for many items, and slow cook over time. Normally, you'd need to get thermometers and constantly check the temperature of the water, or get circulating water baths (that are very expensive) to automate the process. I think this would do very nicely.
I love cooking, and my GF thinks most of my food is better than the resturaunts we go out to eat at. I could see buying this - it's a nice bonus over the induction cooking I'm already lusting after.
You can patent a mutation with a purpose: to create a treatment, etc (as I said above.)
Genes haven't been patented since the late 90's. Before the human genome was completed, I worked @ the whitehead (the center that did the largest amount of the sequencing work.) At first, software was written to auto-patent sequence (to essentially copyleft it), but that work was stopped in the late 90's when it was no longer required.
Maybe this slipped in, as it's a few years before hand. If the courts rule that you can no longer patent something, do existing patents still stand regardless? I'm not an IP lawyer, so I couldn't tell you, but I'd say if someone had money to really pursue the gene, they'd crush that patent in court, right-quick.
As to trade secrets: they should have just patented the mutation with the purpose of making a theraputic out of the target, or a detection method. Those are reasonable. If you spend that much time finding a single mutation that has a strong enough effect to make a drug target, then go ahead and reap the rewards.
By the way, work we're currently doing is at the whole genome scale level, so expect papers in the next few years to say: "We looked at 99% of mutations in the human genome, and found polymorphisms that increase risk to disease . See figure one for details.". Many genes generally effect a trait, so patenting one doesn't give you a monopoly on all cures for a disease.
Re:I worked for a bioinformatics company ...
on
Biotech Data Mining
·
· Score: 1
Actually, you can 'sample DNA' (I think you mean genotype) at any one of the 3 billion bases in the genome, provided you can develop an assay.
The 500K you refer to could be from one of the (now standard) affymetrix 500K chips that have a standardized set of variants to test. They exclude a number of other markers, but via indirect evidence (knowledge of haplotype structure in the population), you can test other areas as well, and increase your coverage of the whole genome.
I'll agree that the statistics are complicated. What you need are ways to define your patients that are matched to each other, except for the phenotype in question you want to test. I find that this information is usually lacking (hell, even the researchers are often saying "I wish that I had more phenotypic data available for my cohort!")
What I'm interested in is looking at their algorithms, and seeing if there's anything I can use for our own work. If they have some well implemented tests that I don't, then that saves me the effort of developing them myself.
Alternative hypothesis - Luxury cars (BMW, etc) will have more reported errors, because users expect their cars to be perfect, while users of non-luxury cars will not complain (or pay to get fixed, and thus be reported) minor problems.
I'm thinking of people I know that have very nice cars, and they must be perfect, or are brought into the dealer. I'm comparing that to my own personal experience of people who don't have 'luxury' cars, who don't care if there's a slight problem with the dashboard plastic, etc.
Any idea how the alternate hypothesis works out? I'm simply curious to see if this is the reliability of the cars, or a reporting bias.
You can't patent a gene. That patent would not hold up in court. Myriad Genetics probably patented the mutation on a gene, with a patent on using that mutation for detection of increased risk of breast cancer, or to develop a theraputic agent.
I work in acedemia (Broad institute), and work on association study work. People follow up linkage results all the time. There's nothing immoral about that, or about the work that is needed to gather a large cohort of patients to do an association study properly. Labs follow up other labs work all the time.
I suggest you get a different example for your bogus patents...
There are lots of other very interesting and handy APIs out there for computational work. I do bioinformatics, and can rely heavily on work other people have already done, and done well.
It's not more accurate, but rather "almost as accurate". 4 errors per article compared to britanica's 3.1 errors per article. And this is only across a sampling of 50 scientific articles, not across the entire spectrum of wikipedia.
Someone obviously didn't read the Nature paper (or even the link posted)!
Add me to your minority, as well as the 30+ programmers that I work near/with.
We're all writing data analysis software (genetics), and testing on small boxes (laptops), and moving to a clustered linux enviornment for bigger jobs. I don't have to change anything to get my code working everywhere. As far as I know, only heavy-client GUIs have any problem at all.
Show me the report, etc that shows how many of each server are running. Where's your quote for the market share?
ALL the 2jee servers I see running are free, across a bunch of different places I work at (MIT, Harvard, etc.) Yes, these are production class enviornments, servicing many people both on and off site, with very complex apps.
What if you get a movie's plot, and the guy next to you doesn't? Are you then too smart, or is the movie shit?
I don't know that a movie has to reach everyone to be a good movie, just like not everyone will understand certain music or art. Does everything have to be dummed down to the lowest common denominator?
Of course, theoretically, nothing is random. We just need to find the rigth angle to view it from, so that it looks the same as it was before. ----
You just need to figure out what the null model is, and then determine if your data fits that model or not.
If I flip a coin a million times, and the coin is unbiased, you'd expect 500,000 heads. If the number of heads is not 500,000, you can make a guess as to what the likelyhood of the series of events you just observed was.
When I read in a paper how someone works with their statistics, I can get an idea of what sort of effects they were looking at, and I can begin to determine what the likelyhood of their seeing the effect was.
I like to see hard numbers (and they're required for publication). They might not make sense in a newspaper, but then, oftnen the people in charge of writing are not equipped to interpret the results.
Yeah, I know, all you can do is try to show that all alternative hypothesis are less believable.
But then, that's as far as science can ever go. At some point, enough overwhelming evidence may convince you that you're right, until something shows up to tell you otherwise.
What those scientists did is try to think of all the other reasons that the number of bacteria might also be decreasing. Then, test all those other hypothesis. When all those alternates don't pan out, and this one does (penicillin vs. non penicillin results in an effect of X reduction of bacteria over N number of tests with a confidence of W) then you can believe your hypothesis.
I take it very few people on Slashdot DO science for a living. I have a paper in nature genetics this month (well, on line, it'll be in print in january), does that count as 'doing science'?
I think you're looking for the phrase "correlation does not imply causality". However, there's a lot of 'statistical based research' where statistics tell you how believable the evidence is that some event occurring is not by chance.
Showing that things are not random can show causality. You need to just look at the data (and many sets, sometimes) from different angles to rule out any alternative hypothesis.
How do you think any genetic studies are conducted?
I had a horrible time getting my player to work, and eventually got rid of it.
I want my player to be an appliance that "just works", not a hacker's fest where I get to debug beta hardware and software, and if i'm lucky, i get to hear some music.
The hardware for the Neuros audio player was horrible, as of a year ago when I purchased one. The radio didn't pick up any stations, the 20G HD model was huge, and the HD on mine wouldn't even spin up 1/2 the time. The batteries were known for dying out, or not charging enough to run the player.
It was a hacker's device, and far from "just working". I love playing with toys and getting things to work, but not my MP3 player. I just want it to start when I hit start, and play some music.
When the player doesn't even do that, you've lost me as a customer.
Anyone want to buy my old one, cheap? $20 + shipping? If I still have it, I'd be happy to get rid of the thing. I'm happy with my rio karma (which actually works!)
Do yourself a favor and look at David Reich's papers on Admixture mapping. That might be what you're referring to, or it might not. I'd agree with you that there's a lot of population structure (and substructure!) that most studies don't take into account.
You can use that (admixture mapping), or you can try to reduce that.
By the way, those 80-1000 SNP models are becomming very old school (if you're using a candidate gene approach, the one thing we know about candidate gene approaches is that we're horrible about picking the right genes.) We're starting to get data back on our Affy 500,000 SNP chips (actually 2 250K chips) done across thousands of individuals. The real problem is processing all that data...but the next year or so is going to be VERY exciting for a bunch of different initiatives like diabetes (type II), bipolar disorder, etc.
I suggest that you a) Read the paper. b) Read the followup papers that also discuss in more detail how to use this data to perform analysis.
Q#2: If you're interested in how well this data transfers over to people from other populations, then read the "tag transfer" paper, which should be out in a month or so. Paul de Bakker will have a paper comming out that studies how to apply this data to other populations. The quick summary: you can use this data for other populations. By studying groups like Yoruba (african) populations, you're capturing the most human variation that we all derive from. After the population bottleneck (40,000 years ago), we split into multiple groups, so we included european descent (utah, it's complicated, it's mormons), and chinese and japanese samples. That gets MOST variation.
The new hapmap has already been release at 5x the density.
#3 See answer #2, but yes, you can find MOST of the variation. With the current map, you can indirectly measure 90% of variation in humans. See a paper by Itsik Pe'er for more information.
#4 No. When you have highly correlated R^2 values, and you have more than one, you can predict things. You also probably don't understand how mutations are organised on haplotypes, and how these ancestral chuncks of DNA essentially keep the same set of bases together. Yes, these haplotypes are broken down over time (the african ones are smaller than the european ones, again because the effective population size is smaller in europe and because of the bottleneck event), but still allow for prediction.
I gotta wonder who you work for...I spend a fair amount of my time at the Broad, where most of this research has gone on, so I might have quite a bit more insight into it than outsiders (having seen the work as it was developed, having talked to the researches, etc.)
You can measure SNPs that are undetected by this method by measuring haplotypes that those rare snps might occur on. Once you see an effect, then you can home in on the causal SNP.
Something that's interesting about your statement: if you look at a very rare SNP (less than 1%, for example), then you have very little power to see an effect on disease for this SNP. By definition, 99% of your sample size is not contributing power to your study. Thus, you can't statistically find effects unless you have a) massive sample size or b) massive mendelian effects.
The goal of this data is to study COMMON polymorphisms. That's why it's callled the common variant hypothesis.
I think the essential bonus of the haplotype is that you can infer data without directly measuring it. According to people at the Broad (where David Altschuler hangs a lot of the time), with the 5K map (1M snps), you can effectively measure 90% of the human variation that exists (about 10M snps).
The haplotypes let you determine that if SNP 1 is an A, then SNP 2 is always a G, etc. Ok, for people who like things more technical, the R^2 values of a lot of the snps are high, so they can use these snps, or sets of snps, to predict the values of hidden snps.
This means that disease effects can be measured, even if the causal snp was not measured (as said by the parent post.)
Yeah. My paper was just accepted to nature genetics, and uses the hapmap as one of the data sets. Interesting that they claim it's finished now, when I had the 5K data back in may...
Ok, so this is to reflect the nature paper that is just comming out now, where a variation was looked for every 5,000 bases. The new map is 5x as dense (every 1k bases), and was released on the 10-24-05. The new map should provide a lot more resolution for interesting questions.
Funny, David Altschuler is my former boss, and is one of the heads of the project. Nice guy - and brilliant. I attend his meetings all the time, and he's a fun guy to work with. I'm currently working as the primary computational programmer on another one of his projects...
Slashdot Mirror
You know, you can't get a stove that puts out BTUs like a professional chef's stove. Induction cooking allows you to achieve those kinds of heat outputs (as it only heats the cookware, not the stove, walls neat the stove, etc) in your own kitchen.
That's why I want induction cooking - I can't stir fry food in a wok nearly as well as I'd like with my wimpy consumer stove. As for RFID to control temperature - that might be really useful interested in holding food at a certain temperature to cook, like in vaccume cooking (sous vide). In that case, you don't cook food above 175 degrees for many items, and slow cook over time. Normally, you'd need to get thermometers and constantly check the temperature of the water, or get circulating water baths (that are very expensive) to automate the process. I think this would do very nicely.
I love cooking, and my GF thinks most of my food is better than the resturaunts we go out to eat at. I could see buying this - it's a nice bonus over the induction cooking I'm already lusting after.
You can patent a mutation with a purpose: to create a treatment, etc (as I said above.)
Genes haven't been patented since the late 90's. Before the human genome was completed, I worked @ the whitehead (the center that did the largest amount of the sequencing work.) At first, software was written to auto-patent sequence (to essentially copyleft it), but that work was stopped in the late 90's when it was no longer required.
Maybe this slipped in, as it's a few years before hand. If the courts rule that you can no longer patent something, do existing patents still stand regardless? I'm not an IP lawyer, so I couldn't tell you, but I'd say if someone had money to really pursue the gene, they'd crush that patent in court, right-quick.
As to trade secrets: they should have just patented the mutation with the purpose of making a theraputic out of the target, or a detection method. Those are reasonable. If you spend that much time finding a single mutation that has a strong enough effect to make a drug target, then go ahead and reap the rewards.
By the way, work we're currently doing is at the whole genome scale level, so expect papers in the next few years to say: "We looked at 99% of mutations in the human genome, and found polymorphisms that increase risk to disease . See figure one for details.". Many genes generally effect a trait, so patenting one doesn't give you a monopoly on all cures for a disease.
You misspelled plagiarise.
Actually, you can 'sample DNA' (I think you mean genotype) at any one of the 3 billion bases in the genome, provided you can develop an assay.
The 500K you refer to could be from one of the (now standard) affymetrix 500K chips that have a standardized set of variants to test. They exclude a number of other markers, but via indirect evidence (knowledge of haplotype structure in the population), you can test other areas as well, and increase your coverage of the whole genome.
I'll agree that the statistics are complicated. What you need are ways to define your patients that are matched to each other, except for the phenotype in question you want to test. I find that this information is usually lacking (hell, even the researchers are often saying "I wish that I had more phenotypic data available for my cohort!")
What I'm interested in is looking at their algorithms, and seeing if there's anything I can use for our own work. If they have some well implemented tests that I don't, then that saves me the effort of developing them myself.
Alternative hypothesis - Luxury cars (BMW, etc) will have more reported errors, because users expect their cars to be perfect, while users of non-luxury cars will not complain (or pay to get fixed, and thus be reported) minor problems.
I'm thinking of people I know that have very nice cars, and they must be perfect, or are brought into the dealer. I'm comparing that to my own personal experience of people who don't have 'luxury' cars, who don't care if there's a slight problem with the dashboard plastic, etc.
Any idea how the alternate hypothesis works out? I'm simply curious to see if this is the reliability of the cars, or a reporting bias.
You can't patent a gene. That patent would not hold up in court. Myriad Genetics probably patented the mutation on a gene, with a patent on using that mutation for detection of increased risk of breast cancer, or to develop a theraputic agent.
I work in acedemia (Broad institute), and work on association study work. People follow up linkage results all the time. There's nothing immoral about that, or about the work that is needed to gather a large cohort of patients to do an association study properly. Labs follow up other labs work all the time.
I suggest you get a different example for your bogus patents...
There are some really nice APIs out there for computational work.
For example, those folks over at CERN don't seem to do any computational work, but when they have to, they like to use this:
Open Source Libraries for High Performance
Scientific and Technical Computing in Java
http://hoschek.home.cern.ch/hoschek/colt/
There are lots of other very interesting and handy APIs out there for computational work. I do bioinformatics, and can rely heavily on work other people have already done, and done well.
It's not more accurate, but rather "almost as accurate". 4 errors per article compared to britanica's 3.1 errors per article. And this is only across a sampling of 50 scientific articles, not across the entire spectrum of wikipedia.
Someone obviously didn't read the Nature paper (or even the link posted)!
If by your ass, you mean IBM for one page, and websphere advisor for the other, then I'd say your ass is looking huge right now.
Now, find some quotes on how cigarrets aren't so bad on a tobacco website for me...because that was just pathetic.
Add me to your minority, as well as the 30+ programmers that I work near/with.
We're all writing data analysis software (genetics), and testing on small boxes (laptops), and moving to a clustered linux enviornment for bigger jobs. I don't have to change anything to get my code working everywhere. As far as I know, only heavy-client GUIs have any problem at all.
-Jim
Show me the report, etc that shows how many of each server are running. Where's your quote for the market share?
ALL the 2jee servers I see running are free, across a bunch of different places I work at (MIT, Harvard, etc.) Yes, these are
production class enviornments, servicing many people both on and off site, with very complex apps.
Get a Canon 17-85mm f/5.6 EF-S IS. I think they go for about $500. With the multiplier, that's about what you're looking for.
What if you get a movie's plot, and the guy next to you doesn't? Are you then too smart, or is the movie shit?
I don't know that a movie has to reach everyone to be a good movie, just like not everyone will understand certain music or art. Does everything have to be dummed down to the lowest common denominator?
Of course, theoretically, nothing is random. We just need to find the rigth angle to view it from, so that it looks the same as it was before.
----
You just need to figure out what the null model is, and then determine if your data fits that model or not.
If I flip a coin a million times, and the coin is unbiased, you'd expect 500,000 heads. If the number of heads is not 500,000, you can make a guess as to what the likelyhood of the series of events you just observed was.
When I read in a paper how someone works with their statistics, I can get an idea of what sort of effects they were looking at, and I can begin to determine what the likelyhood of their seeing the effect was.
I like to see hard numbers (and they're required for publication). They might not make sense in a newspaper, but then, oftnen the people in charge of writing are not equipped to interpret the results.
Yeah, I know, all you can do is try to show that all alternative hypothesis are less believable.
But then, that's as far as science can ever go. At some point, enough overwhelming evidence may convince you that you're right, until something shows up to tell you otherwise.
What those scientists did is try to think of all the other reasons that the number of bacteria might also be decreasing. Then, test all those other hypothesis. When all those alternates don't pan out, and this one does (penicillin vs. non penicillin results in an effect of X reduction of bacteria over N number of tests with a confidence of W) then you can believe your hypothesis.
I take it very few people on Slashdot DO science for a living. I have a paper in nature genetics this month (well, on line, it'll be in print in january), does that count as 'doing science'?
I think you're looking for the phrase "correlation does not imply causality". However, there's a lot of 'statistical based research' where statistics tell you how believable the evidence is that some event occurring is not by chance.
Showing that things are not random can show causality. You need to just look at the data (and many sets, sometimes) from different angles to rule out any alternative hypothesis.
How do you think any genetic studies are conducted?
Please mod this up.
I had a horrible time getting my player to work, and eventually got rid of it.
I want my player to be an appliance that "just works", not a hacker's fest where I get to debug beta hardware and software, and if i'm lucky, i get to hear some music.
The hardware for the Neuros audio player was horrible, as of a year ago when I purchased one. The radio didn't pick up any stations, the 20G HD model was huge, and the HD on mine wouldn't even spin up 1/2 the time. The batteries were known for dying out, or not charging enough to run the player.
It was a hacker's device, and far from "just working". I love playing with toys and getting things to work, but not my MP3 player. I just want it to start when I hit start, and play some music.
When the player doesn't even do that, you've lost me as a customer.
Anyone want to buy my old one, cheap? $20 + shipping? If I still have it, I'd be happy to get rid of the thing. I'm happy with my rio karma (which actually works!)
Do yourself a favor and look at David Reich's papers on Admixture mapping. That might be what you're referring to, or it might not. I'd agree with you that there's a lot of population structure (and substructure!) that most studies don't take into account.
You can use that (admixture mapping), or you can try to reduce that.
By the way, those 80-1000 SNP models are becomming very old school (if you're using a candidate gene approach, the one thing we know about candidate gene approaches is that we're horrible about picking the right genes.) We're starting to get data back on our Affy 500,000 SNP chips (actually 2 250K chips) done across thousands of individuals. The real problem is processing all that data...but the next year or so is going to be VERY exciting for a bunch of different initiatives like diabetes (type II), bipolar disorder, etc.
You look at linkage disequilibrium between markers.
If you need a lot more information than that, I suggest either a book, or journal articles.
I suggest that you
a) Read the paper.
b) Read the followup papers that also discuss in more detail how to use this data to perform analysis.
Q#2: If you're interested in how well this data transfers over to people from other populations, then read the "tag transfer" paper, which should be out in a month or so. Paul de Bakker will have a paper comming out that studies how to apply this data to other populations. The quick summary: you can use this data for other populations. By studying groups like Yoruba (african) populations, you're capturing the most human variation that we all derive from. After the population bottleneck (40,000 years ago), we split into multiple groups, so we included european descent (utah, it's complicated, it's mormons), and chinese and japanese samples. That gets MOST variation.
The new hapmap has already been release at 5x the density.
#3 See answer #2, but yes, you can find MOST of the variation. With the current map, you can indirectly measure 90% of variation in humans. See a paper by Itsik Pe'er for more information.
#4 No. When you have highly correlated R^2 values, and you have more than one, you can predict things. You also probably don't understand how mutations are organised on haplotypes, and how these ancestral chuncks of DNA essentially keep the same set of bases together. Yes, these haplotypes are broken down over time (the african ones are smaller than the european ones, again because the effective population size is smaller in europe and because of the bottleneck event), but still allow for prediction.
I gotta wonder who you work for...I spend a fair amount of my time at the Broad, where most of this research has gone on, so I might have quite a bit more insight into it than outsiders (having seen the work as it was developed, having talked to the researches, etc.)
You can measure SNPs that are undetected by this method by measuring haplotypes that those rare snps might occur on. Once you see an effect, then you can home in on the causal SNP.
Something that's interesting about your statement: if you look at a very rare SNP (less than 1%, for example), then you have very little power to see an effect on disease for this SNP. By definition, 99% of your sample size is not contributing power to your study. Thus, you can't statistically find effects unless you have a) massive sample size or b) massive mendelian effects.
The goal of this data is to study COMMON polymorphisms. That's why it's callled the common variant hypothesis.
And yes, I do work with these people.
I think the essential bonus of the haplotype is that you can infer data without directly measuring it. According to people at the Broad (where David Altschuler hangs a lot of the time), with the 5K map (1M snps), you can effectively measure 90% of the human variation that exists (about 10M snps).
The haplotypes let you determine that if SNP 1 is an A, then SNP 2 is always a G, etc. Ok, for people who like things more technical, the R^2 values of a lot of the snps are high, so they can use these snps, or sets of snps, to predict the values of hidden snps.
This means that disease effects can be measured, even if the causal snp was not measured (as said by the parent post.)
Yeah. My paper was just accepted to nature genetics, and uses the hapmap as one of the data sets. Interesting that they claim it's finished now, when I had the 5K data back in may...
Ok, so this is to reflect the nature paper that is just comming out now, where a variation was looked for every 5,000 bases. The new map is 5x as dense (every 1k bases), and was released on the 10-24-05. The new map should provide a lot more resolution for interesting questions.
Funny, David Altschuler is my former boss, and is one of the heads of the project. Nice guy - and brilliant. I attend his meetings all the time, and he's a fun guy to work with. I'm currently working as the primary computational programmer on another one of his projects...