Matlab r2017a

The optical density was measured in triplicate six times a day – about every two hours during illuminated conditions – at 750 nm (OD₇₅₀), using a single‐beam spectrometer (Genesys 10S UV‐VIS, Thermo Fisher Scientific Inc., Waltham, USA) with an optical path length of 10 mm. In order to correlate OD₇₅₀ with the cell dry weight concentration, one 5 mL sample of the microalga suspension was taken twice a day. The cells were centrifuged (10 min, 15 000 g), (Rotixa 50RS, Hettich, Tuttlingen, Germany) washed with deionized water, centrifuged again (Mikro 20, Hettich, Tuttlingen, Germany) and dried for 48 h in pre‐dried reaction tubes. OD₇₅₀ and cell dry weight concentrations (c_X) were linearly correlated for each batch process.
The specific growth rates μ were estimated each day during exponential growth, applying the Matlab R2020a function ‘fit’ of the curve fitting toolbox to identify the parameter μ of Equation 1.
$c_x\left(t\right)={c_{x,}}_0·e^{\left(\mu ·t\right)}$ The 95% confidence interval for the growth rate was determined using the function ‘confint’ (Matlab R2017a, Mathworks, Nattick, Massachusetts, USA).

Different light attenuation models like Beer‐Lambert or Reynolds and Pacala [43] were compared (data shown in Supporting Information). A modified Lambert‐Beer model gave the best results for estimating light attenuation caused by D. salina suspended in the flat‐plate gas‐lift photobioreactors (see Table 2).
$I\left(t\right)=I_0·e^{-\epsilon ·\sqrt{c_x}·l}$
I(t) is the measured light transmission at the light‐averted side of the photobioreactor, I₀ the incident photon flux density, ε the specific extinction coefficient, c_X the time dependent cell dry weight concentration and l the length of the light path in suspension (20 mm).
As the incident photon flux density is not constant, a relative photon flux density I_relative was introduced as quotient of transmission and incident photon flux density.
$I_{\mathrm{rela}\mathrm{tive}}=\frac{I\left(t\right)}{I_0}=e^{-\epsilon ·\sqrt{c_x}·l}$ The specific extinction coefficient ε was estimated applying the Matlab R2020a function ‘fit’ of the curve fitting toolbox. The 95 % confidence interval of the specific extinction coefficient was determined using the function ‘confint’ (Matlab R2017a, Mathworks, Nattick, Massachusetts, USA).

Quantitative analysis on local compressive strains were reported as mean, median, and inter-quartile range between 25th and 75th percentile of the data on violin plots. Two-tailed Pearson correlation analysis was performed between microstructural parameters (i.e., TMD, BV/TV, and Tb.Th) and mechanical properties (i.e., stiffness and mean local compressive strain) for all six specimens. Additionally, two-tailed Pearson correlation analysis was done between the mean local strains at the most (20% of the BV highly strained) and least (remaining volume) strained regions at the maximum applied compression and microstructural parameters. Both regions within the same specimens were considered related, thus the non-parametric Wilcoxon signed rank test (α = 0.05) was used to compare strain magnitude, trabecular thickness, and tissue mineral densities. All statistical analyses were done using Matlab software (Matlab R2017a, The MathWorks Inc., Natick, MA, USA).

Functional images were preprocessed using the CONNv17 FC toolbox (43 (link)), implemented in MatlabR2017a (The MathWorks Inc., Natick, Massachusetts, USA). The preprocessing pipeline included realignment, denoising of motion artifacts, and head motion (aCompCor (44 (link)), segmentation, coregistration to each participant's anatomical scan, normalization to an age-specific T1 template for pediatric studies (45 (link)), re-sliced to a 2-mm isotropic resolution in MNI space, and smoothing using a Gaussian kernel of 6 mm full-width at half-maximum [FWHM]). Additional steps after denoising included band-pass filtering of the BOLD time series (between 0.008 and 0.09 Hz) and linear detrending. After the motion artifacts and head motion detection, 34 children were excluded for excessive motion (which is to say, those with <4 min of data were excluded) (42 (link)).

The correlation of macrophage-related genes and glycolysis-related genes in a prostate cancer cohort was computed using level 3 gene expression estimates from the RNA-Sequencing in the TCGA PRAD database, extracted and hosted by Firehose DB (BROAD Institute, https://gdac.broadinstitute.org/). The expression estimates were derived using RSEM (Accurate transcript quantification from RNASeq) method.^{26 (link)} In Fig. 1a and S1A, the original level 3 Illumina HiSeq RNAseqV2 RSEM gene-level normalised mRNA expression data for TCGA PRAD was downloaded from the TCGA data portal in March of 2016 and log2 transformed, log2(x + 1). The 333 primary prostate tumours and associated clinical information, including reviewed Gleason score, were retrieved from the TCGA PRAD333 publication.²⁷ Box and scatter plots were generated in MATLAB R2017a (MathWorks Inc.).Fig. 1

Macrophage infiltration correlates with MCT4 expression. Correlation between CSF1R/CD206 and MCT4 mRNA expression in early-stage patients (Gleason score = 3+3) and advanced prostate (Gleason score = 3+4, 4+3 and ≥8) retrieved from TCGA PRAD333: R = 0.719 and 0.643, respectively

We performed the simulation and the data analysis using Matlab R2017a (Mathworks, Inc.).