The Principal Component of Odor

Noam Sobel of the Weizmann Institute gave a talk on olfaction at UniGe this week which was pretty cool. Before diving into his talk, let's look at the big picture.

Olfaction overview

The defining question of olfaction is, how are odors encoded in the brain? (Good God that question sounds obvious.)  Is the odor representation sparse or dense: do odours activate many neurons or few; and do single neurons respond to many odours or few?  What are the spatiotemporal characteristics of the reponse; does it evolve over seconds or milliseconds; is there an ensemble code?

To answer these questions, olfaction researchers have performed the expected experiments: they've recorded from a cell, while presenting a gamut of odors; they've imaged/recorded from dozens of cells, and presented a panoply of odors; and they've sliced the neural responses into bins to see what that tells them.  And they all now agree that the odor code is either sparse or dense, and that the spatiotemporal properties of the response may or may not matter.

Yet no one has addressed one simple question, how are the primary features of odors and perception related?  This question is what Sobel's talk was about, and what he addressed in two papers. (I should mention here that he's written more than two papers on this subject, but these are the two I read after his talk, and they seem to cover most of the important points.)

Structure and perception

The first paper asks, what is the best way to characterize an odor?  From the chemical perspective you can describe odor molecules in terms of chemical bonds, aldehyde groups, etc.  And from the human perspective you can describe odors in familiar terms like "nutty" or "rancid."  What Sobel's group asked whether there was a correlation between our cognitive descriptors and the chemical properties of odors.

To do this he performed principle component analyses* (PCA) on two datasets.  The first was on an old dataset of perceptual descriptors of odors made by a guy named Dravnieks in the 80s.  Dravnieks sent odor samples to ~150 odor experts (e.g. perfumers), and asked them to rate the compounds on ~150 descriptors from 1-5 (e.g. almond, lemon, sweet, rotten egg).  What Dravnieks discovered was that all humans had the same general perception of odor, across cultural and genetic boundaries, which was a big finding in the 80s.

* Principal component analysis is basically used to find correlations between variables. The example Sobel used is that the height and weight of a person are both correlated with our idea of "size."  If I told you someone was big, you would be surprised if he was short, or skinny.  There's a decent intro in this pdf. Or you can Google it.

Sobel took these ratings of the odors, ran a principal component analysis, and found that one principal component could explain half the variability of the dataset.  This component was highly correlated with ratings for "fragrant, floral, and sweet," and negatively correlated with "sweaty, rancid, and sickening," (panel A below) so he labeled this component "pleasantness."

A) Labels that correlate with the first principal component. From Khan et al (2007)

Despite what other descriptors you might use for a smell, the most important thing about it is how much you like it.

Sobel then did the same trick on odors' chemical structure.  He analyzed a database of 1500 odors which characterized each odor in terms of molecular weight, charge, number of carbon atoms, etc.  And he found one principal component could describe 30% variability of chemical compounds.  The categorization of this was somewhat more elusive, but this component had some correlation with the weight and density of the molecule, but not very strongly.

(It surprises me how little molecular weight matters in this analysis.  I have only a facile understanding of PCA, but shouldn't weight be extremely well correlated with some principal component, even if it's not the first? The database they used had hundreds of descriptors, many of which may be marginal and simply muddying the dataset.  At the same time, many of these chemical descriptors are obviously codependent, and PC1 picked up a lot of the variance, so what do I know.)

So far we've learned that we perceptually categorize odors like we categorize movies, as good or bad. The most important chemical descriptor of  molecules is not really intuitive.  But what's really cool is Sobel compared the first principal components  of perception and structure against each other, and found that they are correlated!

A) Perceptual PC1 is correlated with structure PC1, but none of the other PCs are correlated. From Khan et al (2007)

Something about a chemical, j'ne sais quoi, is able to predict how much we will like an odor. The paper continues with some behavioural experiments to verify this idea, but I leave that as an exercise to the reader.

Part deux


Given the correlation between our perception of odors, it's possible that our neurons are encoding something similar.  The second paper looks at this by performing a meta-analysis of twelve papers that recorded from olfactory areas while stimulating with odors.  So he did what he do, and performed PCA on the responses, and found that the first PC of the responses was basically the overall neural response (of course, I'm not sure what the other PCs are. Individual cell activity?).

From Haddid et al (2010)

There are some interesting exceptions in Fig. B, H, and I, but those are due to special chemical groups like aromatics or aldehydes.  I do have a question here as to whether he mislabeled the PC, as neural response is highly correlated with detectability; you can't get a neural response to something you can't detect, but that's just semantics.  He did do some controls to make sure odor concentration was not being indirectly encoded here, but that's a detail.

Where it gets cool is when he plots neural response PC1 against a pleasantness rating, and finds that they are correlated:

From Haddid et al (2010)

This means the odors that we find the most pleasant are the ones that activate our neurons the most. It's almost like our brain evolved to let us sense what's important. Unlike the other analyses, the second PC of activity here was interesting, as it was correlated with odor toxicity.  In the talk Sobel described his idea of these components as something like, "The first PC determines whether we should investigate something, and the second determines whether it's ok to eat it."

Thoughts

It's amazing what you can discover when you ask simple questions.  One big caveat to the results is that while all these principle components are correlated, the correlations are not very strong, usually r~0.4-0.5.  This of course cannot account for much of the variability, but considering how diverse the datasets are, it's impressive that anything meaningful can be extracted at all.

The other big issue I have is that it seems to be missing obvious categorizations like floral, or acidic, which have both perceptual and chemical meaning.  These categories are of course small and exclusive, and may not show up in a PCA because they do not correlate over large groups of chemicals/odors, but they seem meaningful.  But, as I said before, I am not an expert on what PCA can bring out of the data, and it seems like Sobel was searching for general principles rather than specific categories.

References

1. Dravnieks A

(1982) Odor quality: semantically generated multi-dimensional profiles are stable. Science 218:799–801


Haddad R, Weiss T, Khan R, et al. Global features of neural activity in the olfactory system form a parallel code that predicts olfactory behavior and perception. The Journal of neuroscience : the official journal of the Society for Neuroscience. 2010;30(27):9017-26. Available at: http://www.ncbi.nlm.nih.gov/pubmed/20610736 [Accessed July 9, 2010].

Khan RM, Luk C-H, Flinker A, et al. Predicting odor pleasantness from odorant structure: pleasantness as a reflection of the physical world. The Journal of neuroscience : the official journal of the Society for Neuroscience. 2007;27(37):10015-23. Available at: http://www.ncbi.nlm.nih.gov/pubmed/17855616 [Accessed April 27, 2011].

Check, check, oops

Six months ago I read Atul Gawande's Checklist Manifesto which extols the usefulness of standardized procedures for routine but complex tasks.  He first illustrates this idea with the example of checklists in the operating room, which have been shown to drastically increase effectiveness and reduce complications.  These surgery checklists include simple verifications like making sure everyone in the room knows each others' names, antibiotics were administered, and double checking that the procedure to be performed.  While these steps are obvious, they are also easy to forget; by following a checklist these routine mistakes are practically eliminated.

In the rest of the book Gawande considers other checklists and their context.  Probably the most famous checklists are airline pilots'.  Prior to the implementation of these checklists, airplane crashes were not uncommon; now flying is the safest way to travel.  Another example is how venture capitalists create checklists when deciding to invest in companies, including things like how leveraged the company is, or reading the footnotes in cashflow statements.  Once the VCs find a red flag, they quit analyzing and move onto the next company.  By using these checklists, they can simultaneously filter companies quicker, and increase their success rate.

As I read the book, I became excited about applying these ideas to science. Science is littered with checklists.  On the simplest level, we use recipes to make solutions. Slightly more complex are the protocols for mini-preps, PCR, and dissections (a terminal form of surgery) which  work fairly reliably.  And beyond that there are the checklists for performing actual experiments or project management.

Starting in a new lab has provided the perfect opportunity to apply these ideas, as I learn all the lab's techniques. Or in other words, I follow and write protocols. Seizing this opportunity, I started a Google docs file, and have written protocols for head-posting animals, viral injections, and multi-electrode recording.

Besides eulogizing checklists, Gawande characterizes effective checklists.  He distinguishes between DO-CHECK lists in which you perform a set of tasks, and then confirm they are all finished; and READ-DO checklists in which you read each step out as you do them (i.e. a protocol).  The checklists I have written mix these two types freely.  I usually start with a "Gather" section, including things getting ACSF or bleach, which if available will smooth out the procedure.  Then each subsequent section is a READ-DO list of what the procedure entails, with occasional stop points to make sure everything is ready for the next step.

Probably the most important part of creating and implementing checklists is that you continually improve them, and by doing so improve your procedure.  I've used my new protocols a few times since inception, and I always keep a pen handy to  write notes, including things I forgot, item locations, or clarifying language.  Probably the hardest parts of these lists are descriptive elements like, "Put a ring of cement around the olfactory bulb," which basically requires a picture of an mouse skull (I just checked, and this is possible in Google Docs, so I should probably start doing this).

This idea of constant revision speaks to the greater truth of becoming skilled at anything: the most important part of improving is that you put attentive effort into improving the process, and thereby improving results.

My Brother John

When I interviewed at Berkeley the first time, I also visited my brother in Mountainview.  He picked me up at the airport, and we went out for Chinese. He's about fifteen years older than me, a brother by another mother.  I hadn't seen him as an independent adult, so the social dynamic was in flux, but that is a story for my personal blog.  I was well dressed (for me) in a corduroy blazer, and oxford shirt.

Rather than stay with him, I got a room in a motel. The room was unusual for a cheap motel: among other things, it had a jacuzzi and two floor-to-ceiling mirrors. I spent the night polishing my talk, and the next morning went downstairs for breakfast, and to meet my brother.

He was late, so I started chatting with the desk clerk, the same guy who checked me in the night before.  He was a gay Latino studying to be a court reporter specializing in Spanish-language cases.  He wanted to help "Mexicans," which sounded wrong to my gringo ears.  While we were talking, he explained why my room had special amenities: the motel was a notorious prostitute hotel, and the police had installed cameras in the hallways to keep track of people. As the clerk, he saw a sliver of people's private lives, and was witness to multiple affairs.

He asked me, "So who are you waiting for?"
"Oh, my brother's going to pick me up.  He's running late."
"That was your brother last night?"
"Yeah... why?"
"Well, when you checked in yesterday, I thought he was your sugar daddy."

I'll take that as a compliment.

Dumb again and knowing it

Starting a post-doc is frustrating.  Two months ago in my graduate lab, I was able to perform experiments by myself, analyze the data quickly, and fit it into a larger picture.  Now I need support to perform routine experiments, know how to prepare data but not analyze it, and have only read enough literature to start the edges of a jigsaw puzzle.

It's like being a first year again, except now I know how frustratingly limited I am.

Two weeks ago I started a post-doc in Alan Carleton's lab at the University of Geneva, which has entailed completely switching my field.  My background was in cellular neuroscience studying AMPA receptor trafficking and synaptic plasticity in slices.  I am now investigating olfaction (and hopefully taste) using multi-electrode recording and channelrhodopsin in vivo. Where I once had a good handle on the literature of AMPA receptor trafficking, I know only the basics of olfaction.  And while I have done some in vivo electrophysiology before, the details of recording from awake animals and analyzing thousands of spikes of data is daunting.

Before I started my post-doc, I was somewhat aware of my ignorance, and optimistically planned that it would take me 3-6 months to get moderately well trained.  Yet now that I am climbing the learning curve, the slope seems steeper than I anticipated, even while by reasonable standards I am doing just fine.  Maybe it just means I am learning that much more.

One thing I've learned is that no one has any idea what serotonin does (boy am I going to regret typing this when I finally stumble on a paper that explains it). Part of my project involves stimulating serotonergic centers, and measuring how that influences olfactory bulb processing. As a grad student I learned that serotonin is a neuromodulator, and that it's involved in depression and drug addiction (Duke had really great graduate training). So I performed a literature search, and found that all the reviews are from psychiatry journals, and the most cited hard neuroscience review is a 1992 review by Barry Jacobs.  As far as I can tell there has not been a single well-cited neuroscience review focusing on serotonin since Jacobs's thorough, but (hopefully) outdated review.

Getting back to serotonin, it's a fascinatingly ubiquitous molecule. It is involved in a multitude of bodily functions, from bowel movement and vascular dilation, to cognition, motor control and more.  There are fifteen receptor subtypes that are expressed throughout the brain and body, and many targets are directly innervated by different receptor subtypes with diametrically opposing effects.  Neural activity in serotonergic nuclei like the dorsal Raphe is correlated with arousal, and these nuclei are completely silent during REM sleep. Some papers have shown serotonin generally depresses sensory processing, and that Raphe neurons are silent during focused sensation, but it's all rather vague.  Despite all the research on serotonin, there is no obvious neural correlate with its activity, not reward, salience, arousal, attention, decision making, nothing.  Which perhaps is for the best given that it's involved in everything.

I'm not too keen about neuromodulation, but it seems like the field is so wide open that if I discover anything, it would be "significant."

An Ideal Post-doc Interview Schedule

I did a LOT of post-doc interviews: ten labs, eight institutes, on three continents.  I've done two in a day, and one in a day and a half.  I've spoken with lab members who barely spoke English, and others who could out-geek me.  And having done all this, I would like to present my ideal post-doc interview schedule.

Before any flights are arranged, there should be a phone interview.  This is in part to make sure there is mutual interest in terms of project and personality.  But more importantly, this builds the groundwork for the in person interview: you can get simple questions out of the way, like what the current projects are, or what types of technology are available.  Once you arrive in person, the conversation can continue at depth, and you can do things that are natural in person, like look at data, or be social.

I like to arrive the morning or afternoon before the interview.  This gives you a day to take the measure of a city, walk around the university and the downtown, use the public transportation, and see who lives there. One of the most common questions  during interviews is, "How do you like living here?" and the best way to answer that is by living there, even for one day out of a hotel.  This also gives you a topic of conversation in the lab, and a chance to relax rather than having just stepped off a plane.  It may be tempting to have dinner with the lab the night before, but this is too much: it only takes one day to evaluate a lab.*

The night before I like to practice my talk one last time in the privacy of my room, so it's fresh for the morning.

The interview itself should generally run from 10AM-5PM.  The late-morning start allows some leeway for jet-lag, and shortens the day so an introverted scientist doesn't get worn out.  To some degree, the order of events during the interview is like a baseball lineup, where what you do matters more than the order, but my itinerary would go like this:

10-11: The science talk.  This lets the PI (and the lab members) know what the interviewee is doing, so you don't waste time during the one-on-one ("What was my project? Well I'm about to give an hour talk on it...")
11-12: Meet with PI, discussing the standard topics.  Hopefully the PI will test the interviewee a bit, and not get lost too much in their data.
12-1: Lunch nearby campus, at a place people actually dine at.  It's critical that either lunch or dinner should be without the PI, for it shows he or she trusts their people, and it lets you pump them for info.  One interview went disastrously awry when the lab members collectively bitched about their PI.
1-4: Lab tour, meeting with people in the lab, and a break. If you can, go get a coffee (or caramels) to get out of the lab for a bit. After hours of talking and being on, one can get tired, so it's nice to let other people lead the discussions.  I also like to get a 30 min break so I can check e-mail and generally veg. out.
4-5: Meet with the PI again to wrap things up.
5: Dinner! Hopefully the PI asked ahead of time if I had any preferences for fare.  I'm somewhat of a foodie, so I probably put too much weight on this.  Seeing the PI interact with the lab members at dinner is a good measure of their relationship.

And done by seven or eight o'clock,  you can decompress in the evening (or even fly out if you want to save time).  That's my ideal interview.

*One of my favourite things to do when traveling is discover new restaurants, and the best website I've found for that is Chowhound.  It allowed me to discover Serious Pie in Seattle and Fatty Crab in NYC among many others.  Message boards like Quarter to Three can help too.