The most recent manuscript I have become involved with is once again showing that our decades-old perceptions of Archaebacteria and Eubacteria being two distinct “Domains” of life has been constructed by thinking that “There is a Tree”.  When you accept that there is no Great Tree Of Life and instead we have an ensemble of genetic goods, perhaps with a bunch of rules for how these genes are spread around (Hint: it is not a tree-like process), then you get to see so much more.

What is so wrong with the protocol that you first reach for the tools of networks and examine the data in a network framework, then, perhaps (if appropriate), you might move on to a tree-based analysis (Hint: You might do this if the data were genuinely tree-like)?  The worst that could happen is that you are not limiting or blinkering yourself.

This is a simple message and this latest paper is demonstrating a simple thing – Archaebacteria is not a monophyletic taxon in the same way that Vertebrata is a monophyletic taxon.  Why? Because Horizontal Gene Transfer, that’s why.

Reference:

Origins of major archaeal clades correspond to gene acquisitions from bacteria.  Shijulal Nelson-Sathi, Filipa L. Sousa, Mayo Roettger, Nabor Lozada-Chávez, Thorsten Thiergart, Arnold Janssen, David Bryant, Giddy Landan, Peter Schönheit, Bettina Siebers, James O. McInerney & William F. Martin. Nature (2014) doi:10.1038/nature13805

Sequence Similarity Networks have been in use for almost 20 years but not everybody is familiar with what they are and how they can be used.

Essentially they are a very simple way of representing evolutionary history. In many ways there are like phylogenetic trees and in many ways they are not. The nodes on a sequence similarity network (SSN) can represent a sequence and the edges connecting those nodes can either be homology edges or perhaps co-occurrence edges.

However the fundamental unit of the entire network is a pair of nodes that are united by an edge.

In this case, node 1 and node 2 are connected, either because some (all) of their sequences are homologous to one another, or because they are both found together on the same DNA or protein sequence.  In the former case, this would be a Sequence Similarity Network and in the latter case, it would be a co-occurrence network.  I will use the terms Network and Graph interchangeably.

The entire graph can be found as a single connected component or there may be multiple connected components in the graph.

Two nodes are in the same connected component if there is at least one path that can be traveled from one node to the other.  Sometimes, this path might have to go through other nodes and some times this path is direct – a path length of 1.

A Clique is a set of nodes that are maximally connected – they are all connected to each other.  In the first connected component above we can see that nodes 3, 4 and 5 are a clique and also that 1 and 2 are a clique and 2 and 3 are a clique.  However, 1, 2 and 3 are not a clique because they are not maximally connected.

This brings us to the notion of a Shortest path.  Unlike a phylogenetic tree, there may be multiple ways of traversing a graph in order to go from one node to another.  Therefore, we are often interested in the shortest path you need to travel, rather than some arbitrary round-about path between two nodes.  There might be more than one shortest path uniting two nodes.

The Degree of a node is the number of edges that connect to this node.  A high degree node is often a very central, important node in a graph.

Nodes can be central in the network or they can be peripheral and that might be a very interesting thing to analyse.  There are a few kinds of centrality measure.  Closeness centrality is a measure of how far away, on average, a node is from the rest of the nodes.  A central node on the graph will be closer than a peripheral node.  Betweenness centrality is a different concept.  Betweenness centrality is a measure of how many shortest paths must go through this node.  Betweenness centrality is interesting because it often picks out nodes that join communities of nodes in a graph.

If recombination never happened then every homologous family would form a kind of clique provided there was no false negative identification of homologies. However we do know that recombination does occur. This means that there are paths relating groups of sequences. Chimeric sequences hold a graph together and facilitate the formation of giant connected components.

There is no way of representing this entire evolutionary history on a single phylogenetic tree.

At the moment we really are using off the shelf software and methods in order to construct these networks, However in future it would be awesome to have software that understood the kind of data we are looking at. It would be great to relate evolution along network edges to mutations and changes in the composition of proteins and nucleotides. I have no doubt this will happen soon.

I was asked a few times about my work and whether I was meeting with a lot of resistance to the idea that the Tree of Life might not be a real thing and we are better off changing the way we think about the evolutionary history of prokaryotes.

The true answer to this question is that for ecologists, this is not really a new idea – prokaryote phenotypes cannot be mapped onto a unique tree, because of horizontal gene transfer and that is not a controversial thing for them.

However, for evolutionary biologists, this is indeed problematic for some.  Many/most people have been brought up on trees.  They feel that their careers depend on trees being the most accurate representation of evolutionary history and they would prefer things to stay that way, whether it is sensible or not.

Probably the most horrible thing was a review we received 2 years ago for the Alvarez et al paper in PNAS.  It was written almost entirely in UPPERCASE letters by somebody that is clearly demented by the prospect that we would not say there is a tree of life.

This is really tragic for science – when you refuse to look at the data because you feel the answer and you have special insight.  But this is the truth – a small number of people don’t want to know.

I believe this is attributable to Paul Arriola: http://www.elmhurst.edu/~bio/arriola/

I found it funny. But also sad, so, so sad.

Speakers range from dull, disinterested, indifferent, uninspiring, tear-my-hair-out-at-its-roots-im-so-bored to baselessly super-enthusiastic or jocose or bizzarre or completely annoying and uninspiring to – just occasionally – truly inspirational and awesome.

In my time, I guess I have spent more than 1,500 hours standing in front of various crowds of people and speaking to them about something or other and much more time sitting in an audience, most of the time wondering what would happen if I just stood up and shouted at the speaker: “I’ve had enough, you dimwit”.

So, naturally, I have a list of things I feel are necessary to think about for any talk to be a success and a list of things that are necessary to avoid to try to stop it from becoming a crushing failure.

Here is my top five list, though I have many more things I might have included.  I don’t know if this blog post contains the best advice ever (it probably does, though), but in any event, please read it, because some day I might have the privilege of sitting through your talk and maybe these guidelines will prevent your from boring the pants off me.

1. Know your audience

Before every talk, I ask the organiser to tell me who the crowd are likely to be.  If it is a lecture in my own university, I try to figure out what the students are likely to already know – I mean, outside of the fastest route to the college bar.  There is nothing more interminable than getting a lecture that is either so low-pitched and basic that it is tiresome or so high-pitched and advanced that it cannot make any sense to the given audience.  We – all of us – build on our existing knowledge, so any lecture must guide the audience from where they are now to where I want them to be, so for that reason I need to know who they are.  The structure of the lecture is completely guided by who the audience is going to be, not on what I personally feel would be the best possible lecture I could give on the subject – the lecture is not about me, it is about the audience.  I know it should be about me, but C’est la vie.

2. Make great slides if possible.

While some might disagree with me and instead prefer to have 1,000 words on every slide in six-point font, I find that the optimal number of words on a slide is 1.  My favourite slides have one word and maybe one picture on the slide.  This means that there is a central theme to what you are going to say at this point and it means that the audience needs to listen to you, rather than spend their time reading an essay on a powerpoint slide.  Naturally, it is not possible to give a talk where every slide has only one word written on it (some, hopefully will have no words), but from time to time it will be necessary to have slides with more information.  However, if you give a talk where every slide is full of text, then at the end of the lecture the audience won’t even remember what you looked like, because they were too busy reading your slides.

3. Tell the audience the problem right at the beginning.

When I started out with my speaking career, I thought it was important to give the background to the work and then after, say, ten minutes, I would tell the audience what problem or issue I was going to address today.  I think this is not the best way to give a talk.  I now lead with the problem, if possible. I try to make it clear why this is a problem, how important the problem is and what the payoff will be if we manage to solve the problem.  By fixing the problem in the mind’s eye of the audience, the rest of the talk will emerge as a way of thinking about this overall problem.  If I want to give the background or historical perspective to the problem, then there I can do this, but only after the audience have a clear idea of the problem.

4. The presentation is a solo-performance act and you are the solo performer.

Some of the worst presentations I have sat through have been given by academics who believe that their presentation should be somewhat similar to the research that they have written in their papers.  They are dry, humourless, monotonic, static and earnest.  In contrast, some of the best talks I have sat through were by feckless, funny, self-deprecating, animated, engaging scientists who were unafraid to let their own personality come through in their presentation style.  The issue is to not let a big personality drown out the science, but if a big personality can use this personality well, they can make the science stand head-and-shoulders above anything else that is happening in the room.  So, my advice is to draw the audience’s attention to you, but only for the reason that this gives you the chance to draw their attention to the work.  Again, this talk is not about you – you are going to annoy the audience if you feel that the best thing for them is to have to sit through an hour of you polishing your ego or practicing your stand-up comedy routine. Vary the pace of the talk, use comedy sparingly, move about the stage, shout, whisper, gesticulate. It’s an act.

5. Finish with a flourish

So many talks just end like a damp squib.  The presenter arrives at the last slide with an expression of relief, mutters thanks to the funding agencies and to the colleagues that made it all possible and then asks the audience do they have any questions.  Everybody looks at their shoes for inspiration and hopes that there are nobody else in the room is going to decide to make everybody sit through this windbag for a moment longer.  don’t be one of these presenters!!  Shout at the audience, tell them why they have just heard a thing of beauty, tell them that they now know something really cool and that their lives are so much better than they were an hour previously. Finish as you started – tell them that there WAS a problem, but that now we have moved on, we know more, maybe even we have solved the problem.  Whatever you do, make sure the audience feel that the talk had a defined beginning, middle and end.

The End!!

For the past 20 years or more, a lot of work has gone into understanding social networks. We have measures for these nodes that include centrality indices. A centrality index is a measure of how central a node is in the network.  A high centrality node is one that might have a lot of other connections, or because of the shape of the graph, the node might lie on a lot of shortest paths between other nodes.  We can find the shortest path through a network from any node to any other node. This has given rise to the phrase 6 degrees of separation to describe the “small world” nature of friendship networks.

Analysing these graphs has revealed a lot of things. For instance we might expect lots of triangles to appear in a graph.  This is called transitivity. If A is friends with B and B is friends with C we might expect B and C to also be friends.

If we want information to travel throughout the network then the best way to do this is to get the information as quickly as possible to a central node.  Central nodes in a network are highly connected, so information can get from a central node to the other nodes quickly.

Given the huge amount of interest in these networks in today’s era of big data, we have a lot of software already developed for network analysis.  This means we can begin to think of analysing other kinds of networks.  All we have to do is borrow the code.

So armed which is new technology, we downloaded a whole bunch of prokaryote genomes and a whole bunch of eukaryotic genomes –  very diverse eukaryotic genomes.

We constructed a bunch of BLAST networks – gene similarity networks and analysed their shapes, their structure, their connectivity.

We found that there were a number of distinctly shaped connected components that were compatible with a fusion origin of eukaryotes.  We found that the eubacterial portion of the graphs showed a lot of variability in a gene number, whereas the archaebacterial portion of the graphs showed be a lot less variability.  In large eukaryotic genomes we found the eubacterial portion was preferentially duplicated.  In small eukaryotic genomes  we found the eubacterial portion was preferentially lost.  The two kinds of genes in the graph did not have the same fate.

We found what we call extended homology connected components that are only compatible with a fusion hypothesis for eukaryotic origins.

All of these results were obtained using network analysis tools.  The same kinds of tools at Facebook uses.

The full results are published here.

The point I am making goal is that we are seeing a merging of engineering algorithms with evolutionary biology.  This is not new of course, because optimisation algorithms have long been used in phylogeny software, for instance.  It will be interesting to see in the future which other algorithms are incorporated into our thinking in evolutionary biology.

It is time for a concillience between engineering and science.  The division is artificial in any case. Why should you feel that engineering is one distinct thing and science is something else quite distinct?  The old-fashioned system of placing these disciplines in different faculties is not something that makes a lot of sense.  I am quite sure there are many many other algorithms, methods, approaches that are commonly used in engineering that could be fruitfully used in evolutionary biology, but they are not known about.

Group selection and kin selection and game theory all came from pure mathematics.  The result has been transformational. Today Joan Strassman was elected to the National Academy of Sciences in the United States.  She has used game theory to try to explain multicellularity.  This discipline all traces back to the importation of methods from mathematics to biology.  This merger was extremely valuable and important.

I anticipate future mergers of technology to be equally important.