PhD in Braunschweig
I've always wanted to do a PhD. I spent years working in the industry and working on my science projects on the side. And now finally, I've made it. I have been accepted for a PhD program in Germany, TU Braunschweig.
This is the start of my research journey and I thought I would document it in this blog. As I travel through space and time, what will happen? What obstacles will I face? What discoveries will I make?
Here is the original PhD advertisement.
"""
ZOOLOGICAL INSTITUTE, TU BRAUNSCHWEIG
Division for Cellular and molecular Neurobiology
(Köster Lab)
Background: The human brain consists of 86 × 109 neurons, whereas the zebrafish larval brain only contains approximately 105. Despite its smaller size, the basic neuronal architecture or bauplan of the zebrafish brain is largely comparable to its human counterpart.
Moreover, not only are the basic building blocks and neural circuitry evolutionary conserved, but in addition larval zebrafish are small and transparent, thereby offering a unique opportunity to observe neuronal activity with high resolution microscopy.
Indeed, the zebrafish was the first vertebrate organism in which neuronal activity throughout the whole brain has been recorded in real time (Ahrens et al., 2013). Although we can now literally see a fish thinking, the obtained data sets are large and require computational methods to gain fundamental insights into the activity patterns of neural networks and their mutual interplay (Haesemeyer et al., 2019).
Job advertisement as PDF
phd_final
Project: We are looking for an enthusiastic PhD student to analyze, model, and possibly predict neural activity. In the PhD project, the candidate will analyze activity patterns of nerve cells throughout the entire brain obtained by advanced imaging techniques such as light sheet microscopy. This requires, first, the identification and segmentation of signals and their 3D reconstruction over time, and, second, the registration of individual brains and their corresponding activity patterns onto a reference brain in order to compare different data sets.
To this end, you will develop deep learning approaches within a Python framework. Since the imaging datasets are usually large (in the terabyte range), GPU-based methods to accelerate data processing routines will be implemented together with existing frameworks such as advanced normalization tools (ANTs). In addition, the candidate should be interested in working in an interdisciplinary and cutting-edge research area together with engineers, biologists, and data scientists.
The goal of the PhD project is to analyze and quantify neuronal activity patterns across the entire brain in response to externally provided stimuli mapped onto a reference brain. This will provide fundamental insights into brain function in health and disease and may ultimately also lead to novel and better treatments of neuropsychiatric disorders such as depression, anxiety, and drug addiction.
You should have a very good knowledge in coding (in particular Python) and computation in general. Ideally, you already have experience in the implementation of deep learning techniques and the simulation and modeling of neuronal networks. Prior knowledge in biology or microscopy is not necessary, but a keen interest in neuroscience is encouraged. You will work in an international team together with neurobiologists and microfluidic engineers; thus solid conversation skills in English is required.
The PhD position is offered in the Köster Lab at the Zoological Institute (https://www.tu-braunschweig.de/en/zoology) in Braunschweig, Germany. The preferred starting date is the 01.07.2020, and the duration is 3 years. The position is part-time suitable, but should be occupied 100% and is aimed to lead to a PhD degree as Dr. rer. nat. at the Life Sciences Faculty of the Technical University (TU) of Braunschweig.
The payment is made according to task assignment and fulfillment of personal requirements to salary group EG 13 TV-L, 65% (approximately: 2.000 €/month net). Applicants from non-EU countries may have to successfully complete a visa process before hiring can take place and are welcomed to apply. The TU Braunschweig aims to increase the share of women in academic positions. Applications from female candidates are very welcome. Where candidates have equal qualifications, preference will be given to female applicants. Candidates with handicaps will be preferred if equally qualified. Please enclose a proof when applying for the position.
Applications: should be sent by e-mail to Reinhard Köster r.koester(at)tu-bs.de, and must contain the following documents.
Motivation Letter including contact information for two references
Curriculum Vitae including complete address, phone number, email address, educational background, language skills, and work experience
copies of bachelor and master degrees and transcript of grades in original language and in English or German translation
additional documents should be provided on request
All documents should be in the PDF format; preferably, please provide the entire application in a single file. Personal data and documents relating to the application process will be stored electronically. Please note that application costs cannot be refunded. Deadline for applications: until position is filled.
Further reading:
Ahrens, M.B., Orger, M.B., Robson, D.N., Li, J.M., & Keller, P.J. (2013) Whole-brain functional imaging at cellular resolution using light-sheet microscopy. Nat Meth, 10, 413–420.
Haesemeyer, M., Schier, A.F., & Engert, F. (2019) Convergent Temperature Representations in Artificial and Biological Neural Networks. Neuron, 103, 1123–1134.e1126
"""
I spent last night googling deep learning methods for image registration. As I understand, a part of the PhD is developing deep learning algorithms for aligning the brains of recorded zebra fish to a reference atlas. Here are some of the things that I found.
There's an open source libarary called Thunder developed by Misha Ahrens' group. Misha Ahrens is is a pioneer of zebra fish whole brain imaging. I found this tutorial on image registration using Thunder http://docs.thunder-project.org/tutorial-registration The tutorial translates an image by a series of translations and then aligns them onto the original image. The API of the library looks exemplary. So easy to use! It's how I want the methods that I will develop to work. Just a few lines of code with all of the complexity hidden away and you've aligned your images. However, the case of translations is a good start, but too simple compared to real world examples where the fish microscope images are warped in non-linear ways as the fish moves in the micro-fluidic chamber.
Here's a convolutional neural network based library https://github.com/voxelmorph/voxelmorph that looks like exactly what I need. It's unsupervised. It's got a very nice tutorial https://www.kaggle.com/adalca/learn2reg/notebook It looks great!
My next step is to warp the fish images that I have (the prof already sent me some data from the lab). They're actually 21 images per fish taken in the z plane. Then I can test the voxelmorph registration algorithm for random warps, starting from translations and affine transformations:-)
Now that I don't work in the industry any more, I can blog my heart out. I can write about everything that I do. The thing is that I really love writing about science and coding. I do it for myself because it's fun and I'm a graphomaniac. When I go for long periods of time without writing, my soul starts to sink into chaos and I become dysfunctional:-D It's so liberating to have freedom of speech and write about whatever you're doing.
:-)
This is the start of my research journey and I thought I would document it in this blog. As I travel through space and time, what will happen? What obstacles will I face? What discoveries will I make?
Here is the original PhD advertisement.
"""
ZOOLOGICAL INSTITUTE, TU BRAUNSCHWEIG
Division for Cellular and molecular Neurobiology
(Köster Lab)
Background: The human brain consists of 86 × 109 neurons, whereas the zebrafish larval brain only contains approximately 105. Despite its smaller size, the basic neuronal architecture or bauplan of the zebrafish brain is largely comparable to its human counterpart.
Moreover, not only are the basic building blocks and neural circuitry evolutionary conserved, but in addition larval zebrafish are small and transparent, thereby offering a unique opportunity to observe neuronal activity with high resolution microscopy.
Indeed, the zebrafish was the first vertebrate organism in which neuronal activity throughout the whole brain has been recorded in real time (Ahrens et al., 2013). Although we can now literally see a fish thinking, the obtained data sets are large and require computational methods to gain fundamental insights into the activity patterns of neural networks and their mutual interplay (Haesemeyer et al., 2019).
Job advertisement as PDF
phd_final
Project: We are looking for an enthusiastic PhD student to analyze, model, and possibly predict neural activity. In the PhD project, the candidate will analyze activity patterns of nerve cells throughout the entire brain obtained by advanced imaging techniques such as light sheet microscopy. This requires, first, the identification and segmentation of signals and their 3D reconstruction over time, and, second, the registration of individual brains and their corresponding activity patterns onto a reference brain in order to compare different data sets.
To this end, you will develop deep learning approaches within a Python framework. Since the imaging datasets are usually large (in the terabyte range), GPU-based methods to accelerate data processing routines will be implemented together with existing frameworks such as advanced normalization tools (ANTs). In addition, the candidate should be interested in working in an interdisciplinary and cutting-edge research area together with engineers, biologists, and data scientists.
The goal of the PhD project is to analyze and quantify neuronal activity patterns across the entire brain in response to externally provided stimuli mapped onto a reference brain. This will provide fundamental insights into brain function in health and disease and may ultimately also lead to novel and better treatments of neuropsychiatric disorders such as depression, anxiety, and drug addiction.
You should have a very good knowledge in coding (in particular Python) and computation in general. Ideally, you already have experience in the implementation of deep learning techniques and the simulation and modeling of neuronal networks. Prior knowledge in biology or microscopy is not necessary, but a keen interest in neuroscience is encouraged. You will work in an international team together with neurobiologists and microfluidic engineers; thus solid conversation skills in English is required.
The PhD position is offered in the Köster Lab at the Zoological Institute (https://www.tu-braunschweig.de/en/zoology) in Braunschweig, Germany. The preferred starting date is the 01.07.2020, and the duration is 3 years. The position is part-time suitable, but should be occupied 100% and is aimed to lead to a PhD degree as Dr. rer. nat. at the Life Sciences Faculty of the Technical University (TU) of Braunschweig.
The payment is made according to task assignment and fulfillment of personal requirements to salary group EG 13 TV-L, 65% (approximately: 2.000 €/month net). Applicants from non-EU countries may have to successfully complete a visa process before hiring can take place and are welcomed to apply. The TU Braunschweig aims to increase the share of women in academic positions. Applications from female candidates are very welcome. Where candidates have equal qualifications, preference will be given to female applicants. Candidates with handicaps will be preferred if equally qualified. Please enclose a proof when applying for the position.
Applications: should be sent by e-mail to Reinhard Köster r.koester(at)tu-bs.de, and must contain the following documents.
Motivation Letter including contact information for two references
Curriculum Vitae including complete address, phone number, email address, educational background, language skills, and work experience
copies of bachelor and master degrees and transcript of grades in original language and in English or German translation
additional documents should be provided on request
All documents should be in the PDF format; preferably, please provide the entire application in a single file. Personal data and documents relating to the application process will be stored electronically. Please note that application costs cannot be refunded. Deadline for applications: until position is filled.
Further reading:
Ahrens, M.B., Orger, M.B., Robson, D.N., Li, J.M., & Keller, P.J. (2013) Whole-brain functional imaging at cellular resolution using light-sheet microscopy. Nat Meth, 10, 413–420.
Haesemeyer, M., Schier, A.F., & Engert, F. (2019) Convergent Temperature Representations in Artificial and Biological Neural Networks. Neuron, 103, 1123–1134.e1126
"""
I spent last night googling deep learning methods for image registration. As I understand, a part of the PhD is developing deep learning algorithms for aligning the brains of recorded zebra fish to a reference atlas. Here are some of the things that I found.
There's an open source libarary called Thunder developed by Misha Ahrens' group. Misha Ahrens is is a pioneer of zebra fish whole brain imaging. I found this tutorial on image registration using Thunder http://docs.thunder-project.org/tutorial-registration The tutorial translates an image by a series of translations and then aligns them onto the original image. The API of the library looks exemplary. So easy to use! It's how I want the methods that I will develop to work. Just a few lines of code with all of the complexity hidden away and you've aligned your images. However, the case of translations is a good start, but too simple compared to real world examples where the fish microscope images are warped in non-linear ways as the fish moves in the micro-fluidic chamber.
Here's a convolutional neural network based library https://github.com/voxelmorph/voxelmorph that looks like exactly what I need. It's unsupervised. It's got a very nice tutorial https://www.kaggle.com/adalca/learn2reg/notebook It looks great!
My next step is to warp the fish images that I have (the prof already sent me some data from the lab). They're actually 21 images per fish taken in the z plane. Then I can test the voxelmorph registration algorithm for random warps, starting from translations and affine transformations:-)
Now that I don't work in the industry any more, I can blog my heart out. I can write about everything that I do. The thing is that I really love writing about science and coding. I do it for myself because it's fun and I'm a graphomaniac. When I go for long periods of time without writing, my soul starts to sink into chaos and I become dysfunctional:-D It's so liberating to have freedom of speech and write about whatever you're doing.
:-)
Kommentaarid
Postita kommentaar