This post may not surprise you if you already read the latest Nature editorial Do scientists really need a PhD? Before we talk about real issues, my first advice for aspiring PhD student - don't read too much into anti-PhD arguments presented here. In fact PhD can be uplifting, accomplishing, stimulating and a life long experience. For me this is really a complex issue, I love doing research as much as I hate the PhD.

I think it's safe to say that non-PhD scientists have always played an important role in scientific eco-system but they hardly get limelight and credit they deserve. What we are seeing here is a new role for non-PhD scientists, data saviors. Currently scientific community is battling with floods of scientific data especially the life sciences domain is generating unprecedented amount of data every day (may be more than 10 terabytes of sequencing data everyday). Community is facing a huge backlog in form of unprocessed data waiting for first hand analysis. I am not in a total aggrement with the interpretation that scientist without PhD are the way of the future, actuality I think that non-PhD researchers will play a key role as data scientists, a role which will be more instrumental in enabling the science than doing the science itself. Its just not about first hand analysis of data, field is being held back by several other key issues such curation and annotation. As data scientists non-PhD research can add tremendous value to the scientific output. What do you think?

Quality and availability of bioinformatics resources is always a matter of great debate. HTTP 404 not found is quite frequent phenomenon for bioinformatics researchers looking to use some published web accessible database or analysis tool. A 4-year follow-up survey on the lack of persistence of bioinformatics resources was published in year 2008 by Jonathan D. Wren suggests that approximately 20% of URLs published in MEDLINE abstracts are now inaccessible, and the most common types of inaccessible content are computer programs (43%) and databases (19%). A similar kind of study was taken by Veretnik et. al where they analysed the availability of web resources published in the Nucleic Acids Research (NAR) Web server issues between 2004 to 2007. They found that a significant number web server goes offline after two years from their data of publication. I suspect the numbers are more higher than what we are seeing here and roughly 50% of published bioinformatics web resources have lifespan of only 2 years or less.

Many people see this availability decay as huge problem in bioinformatics community but I suggest that sudden death of majority of bioinformatics resources simply implies their unpopularity among the potential users and it is part of solution to filter the reliable and quality resources. Jonathan D. Wren used term data tombs for the this kind of web resources, either they are rarely accessed due to there narrow scope

Data tombs, in large part, seem to have resulted from a ‘build it and they will come’ philosophy, which is OK as a means of justifying database creation, but not publication.

or there are major design flaws and lack of user perspective in which case users will fall on alternative option that is less complicated but easier to use. In simple words,

Developers often equate the power of their software with the number of options, but users usually equate the number of options with the number of barriers between them and their results.

One of the major problem with dead web resources is their ghost, aka citations for the published paper which solely describes the resource itself. Nonetheless, people keep citing the dead bioinformatics resources without any check and for all wrong reasons. We definitely need a mechanism to curb these ghost citations.

Apart from the quality and narrow scope there are several other reasons which may affect the accessibility of the web resources. Frequent change of URLs, funding issues, lack of time and resources to maintain the published applications are some of them. Time to time people have suggested various solutions, and two most compelling options are

1. Archiving of web resources in public repositories. This is can be done by a third party very much like Sourceforge, Google Code, Github or Bioconductor. For me the best example here is the BioModels database- a database of published systems biology models - for each release they keep a public dump in the Sourceforge repository. This is something resources developers should consider before they start developing the application especially how they can archive the database or tool. Again in recent past a lot of shops closed due to lack of funding, I think this is something which is inevitability for most of bioinformatics resources currently available except few very big ones such as NCBI, PDB etc. I said before and I will suggest again, rather than depending on funding agencies for a sustainable funding model for bioinformatics resources, project manager should consider these publicly available repositories and make regular dumps for their web resources.

2. Use of PURLs or Persistent uniform resource locators. This is a good alternative for resources with changing URLs. Whenever the URL changes, resources creator has to manually update the PURL to the new URL.

In a latest PLoS Computational Biology paper Bourne et. al argue that widgets centric web resources can change the way the computational biologist currently engage with online database and analysis tools (Thanks to Sushant for pointer to this article which was published very recently). Authors suggest that use of widgets is a compelling idea for long term availability of bioinformatics resources and I also think this can curb the growth of unwanted data tombs. A web widget is nothing but code snippet commonly a JavaScript chunk which can be embedded and executed within any separate HTML-based web page by an end user. Web widgets are around us for quite long time now, any one who maintains a blog knows very well about the widgets and the functionality they bring to the webpage. So what big fuss about widgets in bioinformatics? Well in simple words, if bioinforamtics is an Apple platform (iPhone or iPad) then widgets are just like iPhone application, extending the functionality of bioinforamtics resources in a modular way. Bourne et. al summarize it very well

First, it brings the application to you; it is an example of drop technology (simply drop the application into your Web page) and it facilitates use. You do not have to remember where to go and possibly be faced by a series of complex choices—the widget can offer a simplified interface to a subset of features. Second, and more importantly, assuming the use of widgets takes off, you can customize your own Web page to take advantage of work done by a variety of other scientists each producing widgets. So for example, you could aggregate a variety of remote methods that perform sequence and structure comparison using a variety of widgets from a variety of reputable sources, thereby creating a single point of reference. Taking this a step further, you can create and customize workflows composed of different widgets in a plug and play environment.

Just think about potential implications of this technology, for instance a bioinformatics papers describing a database or analysis tool can be embedded in online version of paper itself, while reading and citing user can check current accessibility or availability of the resources described in the publication and if the resources is not available without further wasting their time to read the whole paper they can just move to alternative option.

Widgets not just to people, but to programs also
If used in its purest form another advantage with widget centric approach will be ability of widgets to communicate with other widgets or programs which is nothing but an API centric model. For that we will need a standards for widget development both on the server side as well as on the client side, may be we need a community wide effort something like The minimum information required for publishing a computer application. I genuinely believe this is a very nice suggestion from Bourne et. al.

Lately Microsoft has released few interesting add-ins for their Office products. Yesterday I wrote about a Word add-in that enables the annotation of scientific documents using bio-ontologies and controlled vocabularies. Today I learned about their latest add-in for Excel, NodeXL, which enables user to create and analyze network visualizations without any hassle. Like the ontology add-in NodeXL is absolutely free and code base is open sourced. NodeXL is primarily designed for extracting, analyzing and visualizing social media networks, in fact it can connect directly to social networking website such as Twitter to import the network data for analysis in Excel. For instance you can import network data of people who have recently tweeted a certain term which might interest you. Coming back to the main motive behind this post, I feel there is nothing which stops us to use NodeXL with biological data. It will be quite easy to extend the NodeXL to import the data from biological databases such as BIND. I know there are plenty of tools already available for the biological networks visualization, interested readers can find a detailed review of major biological network visualization tools by Pavlopoulos et. al. But considering the fact that most of biologist are more familiar with Excel than any other tool out there for network visualization, using NodeXL will make more sense to them. Even in current form itself NodeXL can be used to visualize the biological data, user can import graph data in any major file formats such as GraphML, UCINet, Pajek, and matrix.

Another thing which caught my attention is Microsoft is working with two more bioinformatics extensions, first one is Microsoft Biology Foundation (MBF) a language-neutral bioinformatics toolkit built as an extension to the Microsoft .NET framework and the second one is Biology Extension for Excel which is built on top of MBF and enables a simple and flexible way to work with genomic sequences, metadata, and interval data in an Excel. Both of these products are currently in their beta release and developed by a dedicated Microsoft development team. In current release MBF and its SDK include several features,

Microsoft Biology Foundation:
Core MBF system library (bio.dll).
Core MBF web service library (WebServicesHandler.dll).
Advanced algorithms for use in performing Multiple Sequence Alignment as well as whole genome de novo assembly, included both as core library features as well as examples of how to add modular functionality to the core via the Addin auto-registration mechanism (Addin\PaDeNa.dll, Addin\Pamsam.dll).
Introductory as well as advanced technical documentation to better inform those interested in learning more about the technology and capabilities (Doc).
SDK:
Complete technical reference documentation explaining all available capabilities of the MBF framework (BioDotNet.chm).
Trident activity samples for generating genomics workflows using the Trident workflow workbench (bio.workflow.dll).
IronPython project to show how the MBF library of functional can be used with the Python scripting language (BioDemo.py).
Sequence Simulator, a small sample application which allows for splitting of large sequences into multiple shorter “reads” – intended to replicate what might come off a next-generation sequencing machine.

Despite being a .Net centric project I think MBF is going to offer a serious competition to OBF projects such as BioPerl, BioPython etc in terms of quality, optimization and stability. A final release of MBF is expected around June this year, after that project will be more open for external contribution which currently seems quite restricted.

This is one of many surprises you can always expect from big blues, in fact IBM has always focused to develop the efficient computational algorithms in order to reduce the computational complexity and cost related to the processing of the enormous pools of data. What is new about this latest breakthrough from IBM is the development of energy efficient method to crunch the big data. The mathematical algorithm, developed by IBM's laboratories in Zurich, is tricky and quite general in nature, in fact it can be used to solve several domain specific problems including bioinformatics one. According this IBM news announcement:

In a record-breaking experiment, IBM researchers used the fourth most powerful supercomputer in the world -- a Blue Gene/P system at the Forschungszentrum Julich in Germany -- to validate nine terabytes of data (nine million million or a number with 12 zeros) in less than 20 minutes, without compromising accuracy. Ordinarily, using the same system, this would take more than a day. Additionally, the process used just one percent of the energy that would typically be required.

In a post-Hadoop scenario this work has major implications especially when most of Hadoop implementations are tuned for availability and computational efficiency, not for the energy efficiency.

A latest paper in BMC Bioinformatics describes a Microsoft Word Add-in for ontology recognition. Tool is freely available from Codeplex portal and as prerequisite you will need Microsoft Word 2007. This add-in enables the annotation of scientific documents based on terms that appear in ontologies and controlled vocabularies. I am sure this tool is going to fuel the debate on the article of the future as like everyone else I think that the next generation of scientific articles will be semantic rich and there is urgent need for semantic enrichment of scientific articles. Adding semantic data to scientific articles during the authoring process is one of the option discussed on several occasions but lack of noticeable tools to assist authors in these efforts remained biggest bottleneck in this process. Now this new add-in for one of the most widely used authoring program gives a new opportunity to authors to add semantic data to a scientific document as it is being written.

or provide custom ontologies from the local machine or a remote source. Database ID

recognition can also be activated or deactivated via this panel.

Some of the features currently supported by the tool

*Inline Syntax Coloring of Informative Words
*Built-in Knowledge of Ontologies and Controlled Vocabularies maintained and delivered by NCBO
*Built-in Knowledge of Biological Databases (Protein Data Bank, UniProtKB, NCBI GenBank/RefSeq)
*Automatic Detection of Identifiers
*Custom Semantic Markup

Systems biology is supposed to revolutionise the biology by going beyond reductionism of molecular biology which has dominated biology for decades. The reductionist approach mostly relie on breaking a larger biological system down into pieces, analysing smaller pieces seperately and further determining the connections between them. For nearly 30 years (1970s to the early 1990s) this kind of approach was quite popular and successful in the biology. For a long time a particular type of reductionism – methodological reductionism - has been particularly evident in biology with a mojor focus to understand the physiochemical basis of biological phenomenon. Unfortunately reductionism has its own problems, one of the major concern is the lack of complexity or ability to capture the system level behviour. One of the excerpt from Margaret Drabble's book The Sea Lady portrayes the harsh realities of reductionism,

You can't learn everything from the laboratory, that's what he used to say. The whole is more than the sum of its parts, he told us. The whole behaves differently from the parts, and has different properties. That's what he taught us, and he was right. It's out of fashion to say these days, when we spend our time scrutinizing the interactions of eukaryotic microbes, but it's true, nevertheless. It's still true.

Scientists from Texas A&M University have developed a new algorithm MrsRF (MapReduce Speeds up Robinson-Foulds) for analyzing large collection of evolutionary trees using MapReduce framework. Matthews et. al, have used their MapReduce algorithm to compute all-to-all Robinson-Foulds (RF) distance matrix on multi-core computing platforms. Calculation of all possible Robinson-Foulds distance pairs is a computationally intensive task. The results show that a significant speedup can be achieved using MrsRF compared to the fastest sequential algorithms.

We studied the performance of our MrsRF algorithm on two large biological trees sets consisting of 20,000 trees of 150 taxa each and 33,306 trees of 567 taxa each. Our experiments show that MrsRF is a scalable approach reaching a speedup of over 18 on 32 total cores.

Phase 1 of the MrsRF algorithm. Two mappers and two reducers are used to process the input files.

Apart from speeding up the phylogenetic analysis, this study presents a new type of MapReducible problem where "the size of the input (t evolutionary trees) is much smaller than the size of the output (t × t RF matrix)". Generally in MapReduce implementations the final output is smaller in size than the initial input. Another important thing which authors point out is getting best performance out of MapReduce implementation on a multi-core cluster depends on the cluster configuration. For instance, they tried their problem set with 32 total cores, a 16 nodes by 2 cores (16 × 2) cluster configuration which outperformed 8 × 4, 4 × 8, and 32 × 1 cluster configuration.
Overall their research makes a strong case for using MapReduce framework to design high-performance phylogenetic applications and it can be best for tackling the large evolutionary computational problems such as summarizing the big collections of evolutionary trees. An open-source implementation of MrsRF algorithm is freely available from the Google code.

Dr. Monica Anderson talks about reductionism vs holism in science. She also suggests that artificial intelligence failed because intelligence is too holistic and by its nature it always attracted the hardcore reductionists. One of my favorite quote from this talk is "Science changes the world, sometimes the world changes the science, we are overdue", which is significant in each and every sense. Check out the whole lecture it is worthwhile.

I hope you remember that some time back Microsoft tried to patent clustering phylogenetics methods which was a socking news for the bioinformatics community as community used these methods for a long time without any restriction. Now Microsoft had patented the personal data mining system. According to patent abstract which was accepted just last week

Personal data mining mechanisms and methods are employed to identify relevant information that otherwise would likely remain undiscovered. Users supply personal data that can be analyzed in conjunction with data associated with a plurality of other users to provide useful information that can improve business operations and/or quality of life. Personal data can be mined alone or in conjunction with third party data to identify correlations amongst the data and associated users. Applications or services can interact with such data and present it to users in a myriad of manners, for instance as notifications of opportunities.

The patents includes some heavy weight names from Microsoft including Bill Gates. System tries to answer personal queries such as "What is the best digital camera and where I can find the cheapest one?". What really troubling is the claims like the one below,

Furthermore, as will be appreciated, various portions of the disclosed systems and methods may include or consist of artificial intelligence, machine learning, or knowledge or rule based components, sub-components, processes, means, methodologies, or mechanisms (e.g., support vector machines, neural networks, expert systems, Bayesian belief networks, fuzzy logic, data fusion engines, classifiers . . . ).

Also I am also not sure how a patent was granted purely based on data streams without the looking on the background of algorithm used.

In a latest online article in journal Science Ge et al. report a cryo-electron microscopy structure of a model rhabdovirus, a bullet-shaped vesicular stomatitis virus (VSV). Using their structural data they suggest a mechanism of VSV assembly in which the nucleocapsid spirals from the tip to become the helical trunk.




Architectural representation of the virion trunk, rendered in 3D animation.