Towards an open source "Google Maps"-like interface for neuroanatomical data

Daniel Fürth
Meletis Lab

Lee  lab meeting
22th July 2016

daniel.furth@ki.se

Towards an open source "Google Maps"-like interface for neuroanatomical data

Daniel Fürth
Meletis Lab

Zador lab meeting
11th April 2016

daniel.furth@ki.se

Content:

  • Neuroanatomical Information System (NIS)
    • Input to serotonergic raphe nuclei
    • Cortico-striatal networks
    • Scalable monosynaptic architecture
  • Spatial mapping of single-cell RNAseq data to tissue of origin
  • Multiplexed padlock probes

Neuroanatomical Information System (NIS)

Geographical Information System (GIS)

Web-interface

similar to...

works with...

Surface primitives

Non-Uniform Rational B-Splines (NURBS)

Surface primitives

Surface primitives

www.wholebrainsoftware.org

Tracing the network

Reconstructing brain from sectioned tissue

Pollak et al. 2014

www.mcstan.org

Whole-Brain Reconstruction

Pollak Dorocic et al. 2014

understanding behavior

striatal models

Graybiel, A. et al. (1990) PNAS 

cfos

cocaine

saline

striatal models

Albin, Young & Penney, 1989 

Alexander, DeLong & Stric 1986 

striatal models

Graybiel, A. et al. (1990) 
Jordan, M. I. et. al. (1991). Adaptive mixtures of local experts.

Tracing the network

Tracing the network

DRD2 film
56,710 neurons

Tracing the network

161,294 pooled neurons 

Tracing the network

Tracing the network

Tracing the network

Cocaine induced locomotoric activity

Multiresolution decomposition

Multiresolution decomposition

Multiresolution decomposition

Performance

Multiresolution decomposition

Multiresolution decomposition

Multiresolution decomposition

Segmentation of nuclei

Segmentation of nuclei

Best measurement of c-fos activity?

Monotonic function of IEG activity?

Functional validation

Can be used to segment processes and their direction.

Functional validation

Scalable monosynaptic architecture

Scalable monosynaptic architecture

Stevens (1999)

Scalable monosynaptic architecture

Stevens (1999)

Scalable monosynaptic architecture

Scalable monosynaptic architecture

N_{\text{pre}} \propto N_{\text{post}}^{3/2}
NpreNpost3/2N_{\text{pre}} \propto N_{\text{post}}^{3/2}

Scalable monosynaptic architecture

scRNA-seq

NATURE BIOTECHNOLOGY | COMPUTATIONAL BIOLOGY | ANALYSIS

High-throughput spatial mapping of single-cell RNA-seq data to tissue of origin

Kaia Achim, Jean-Baptiste Pettit, Luis R Saraiva, Daria Gavriouchkina, Tomas Larsson, Detlev Arendt & John C Marioni

scRNA-seq

Allen Brain Reference Atlas

  • Atlas 2007 (manually drawn Nissl):
    • 200 μm thick coronal sections.
  • Atlas 2011:
    • 100 μm both coronal and sagital
  • Atlas 2014 (connectivity avrg template)
  • Atlas 2015 (beginning of june):
    • 10 x 50 μm
  • Registration atlas:
    • 25 x 25 μm
  • Grid expression ISH:
    • 200 x 200 μm MetaIOimage (.raw, .mhd)

 

scRNA-seq

scRNA-seq

Allen Brain Reference Atlas

scRNA-seq

Anatomic Gene Expression Atlas

Lydia Ng,  et al. (2009) Nat. Neuro.

http://mouse.brain-map.org/agea

scRNA-seq

324 cells from cortico-striatal section

scRNA-seq

Our approach

scRNA-seq

Our approach

Multiplexed padlock probes

Mats Nilsson

Multiplexed padlock probes

Reverse transcription

RNase H digestion

Hybridization

Ligation

Rolling Circle Amplification

mFISH

Base calling

Multiplexed padlock probes

Mats Nilsson

Multiplexed padlock probes

Mats Nilsson

Multiresolution decomposition

Thank you!

Concurrency and parallel programming

  • Multi-threaded applications through             .
#include <string>
#include <iostream>
#include <thread>

using namespace std;

//The functions we want to make the thread run.
void task1(string msg)
{
    cout << "task1 says: " << msg;
}

void task2(string msg)
{
    cout << "task1 says: " << msg;
}

//Main loop.
int main()
{
    thread t1(task1, "Task 1 executed");
    thread t2(task2, "Task 1 executed");
    //let main wait for t1 and t2 to finish.
    t1.join();
    t2.join();
}

Rcpp

Dual core

Concurrency and parallel programming

  • Multi threaded applications through             .
#include <string>
#include <iostream>
#include <thread>

using namespace std;

//The functions we want to make the thread run.
void task1(string msg)
{
    cout << "task1 says: " << msg;
}

void task2(string msg)
{
    cout << "task1 says: " << msg;
}

//Main loop.
int main()
{
    thread t1(task1, "Task 1 executed");
    thread t2(task2, "Task 1 executed");
    t1.join();
    t2.join();
}

Rcpp

Parallel computing

  • Parallel computing is extremely simple to implement from R. 

 

# install.packages('foreach'); install.packages('doSNOW')
library(foreach)
library(doSNOW)
cl <- makeCluster(2, type = "SOCK")
registerDoSNOW(cl)

getDoParName()


#matrix operators
x <- foreach(i=1:8, .combine='rbind', .packages='wholebrain' ) %:%
   foreach(j=1:2, .combine='c', .packages='wholebrain' ) %dopar% {
     l <- runif(1, i, 100)
     i + j + l  
   }

R package

  • Why R?
    • Standard data analysis:
      • load some data
      • estimate the density distribution.
      • plot it
xx <- faithful$eruptions
fit <- density(xx)
plot(fit)

R package

  • Why R?
#Line 1: loading
xx <- faithful$eruptions
#Line 2: estimate density
fit1 <- density(xx)
#Line 2: draw 10'000 bootstraps
fit2 <- replicate(10000, {
    x <- sample(xx,replace=TRUE);
    density(x, from=min(fit1$x), to=max(fit1$x))$y
})
#Line 3: compute 95% error "bars"
fit3 <- apply(fit2, 1, quantile,c(0.025,0.975))
#Line 4: plot the estimate
plot(fit1, ylim=range(fit3))
#Line 5: add estimation error as shaded region
polygon(c(fit1$x,rev(fit1$x)), c(fit3[1,], rev(fit3[2,])), col=’grey’, border=F)
#Line 6: add the line again since the polygon overshadows it.
lines(fit1)

What other language can do this in 6 lines of code?

scRNA-seq

Gene specificity

about ~24'000 genes expressed in the brain. 

\text{Let us define the following variables.}
Let us define the following variables.\text{Let us define the following variables.}
c : \text{a unique single cell.} \quad \text{Where: } c \in \{1, ... , C\}, \text{ and } C = 380.
c:a unique single cell.Where: c{1,...,C}, and C=380.c : \text{a unique single cell.} \quad \text{Where: } c \in \{1, ... , C\}, \text{ and } C = 380.
m : \text{a unique single gene.} \quad \text{Where: } m \in \{1, ... , M\}, \text{ and } C = 380.
m:a unique single gene.Where: m{1,...,M}, and C=380.m : \text{a unique single gene.} \quad \text{Where: } m \in \{1, ... , M\}, \text{ and } C = 380.
D : \text{a } C \times D \text{ read count matrix.}
D:a C×D read count matrix.D : \text{a } C \times D \text{ read count matrix.}
D_{c,m} : \text{ normalized number of reads mapped to cell } c \text{ for gene } m.
Dc,m: normalized number of reads mapped to cell c for gene m.D_{c,m} : \text{ normalized number of reads mapped to cell } c \text{ for gene } m.
r_{c,m} : \text{ cell-gene specificity ratio.}
rc,m: cell-gene specificity ratio.r_{c,m} : \text{ cell-gene specificity ratio.}
r_{c,m} = \frac{D_{c,m}}{ s_m }
rc,m=Dc,msmr_{c,m} = \frac{D_{c,m}}{ s_m }
s_{m} = \frac{1}{C} \sum_{i=1}^C D_{i,m}
sm=1Ci=1CDi,ms_{m} = \frac{1}{C} \sum_{i=1}^C D_{i,m}
s_{m} : \text{ gene specificity.}
sm: gene specificity.s_{m} : \text{ gene specificity.}
Made with Slides.com