l4b4r4b4b4
AI / IT-Consultant and Solution provider Full Stack DevOp
Holocene IntelligenceBerlin, Germany
Pinned Repositories
Agentless
Agentlessš±: an agentless approach to automatically solve software development problems
AIDocks
LLM-Training-API: Including Embeddings & ReRankers, mergekit, LaserRMT
ansible-workadventure
arxiv-rss-assistant
Assemblies-of-putative-SARS-CoV2-spike-encoding-mRNA-sequences-for-vaccines-BNT-162b2-and-mRNA-1273
RNA vaccines have become a key tool in moving forward through the challenges raised both in the current pandemic and in numerous other public health and medical challenges. With the rollout of vaccines for COVID-19, these synthetic mRNAs have become broadly distributed RNA species in numerous human populations. Despite their ubiquity, sequences are not always available for such RNAs. Standard methods facilitate such sequencing. In this note, we provide experimental sequence information for the RNA components of the initial Moderna (https://pubmed.ncbi.nlm.nih.gov/32756549/) and Pfizer/BioNTech (https://pubmed.ncbi.nlm.nih.gov/33301246/) COVID-19 vaccines, allowing a working assembly of the former and a confirmation of previously reported sequence information for the latter RNA. Sharing of sequence information for broadly used therapeutics has the benefit of allowing any researchers or clinicians using sequencing approaches to rapidly identify such sequences as therapeutic-derived rather than host or infectious in origin. For this work, RNAs were obtained as discards from the small portions of vaccine doses that remained in vials after immunization; such portions would have been required to be otherwise discarded and were analyzed under FDA authorization for research use. To obtain the small amounts of RNA needed for characterization, vaccine remnants were phenol-chloroform extracted using TRIzol Reagent (Invitrogen), with intactness assessed by Agilent 2100 Bioanalyzer before and after extraction. Although our analysis mainly focused on RNAs obtained as soon as possible following discard, we also analyzed samples which had been refrigerated (~4 ā) for up to 42 days with and without the addition of EDTA. Interestingly a substantial fraction of the RNA remained intact in these preparations. We note that the formulation of the vaccines includes numerous key chemical components which are quite possibly unstable under these conditions-- so these data certainly do not suggest that the vaccine as a biological agent is stable. But it is of interest that chemical stability of RNA itself is not sufficient to preclude eventual development of vaccines with a much less involved cold-chain storage and transportation. For further analysis, the initial RNAs were fragmented by heating to 94ā, primed with a random hexamer-tailed adaptor, amplified through a template-switch protocol (Takara SMARTerer Stranded RNA-seq kit), and sequenced using a MiSeq instrument (Illumina) with paired end 78-per end sequencing. As a reference material in specific assays, we included RNA of known concentration and sequence (from bacteriophage MS2). From these data, we obtained partial information on strandedness and a set of segments that could be used for assembly. This was particularly useful for the Moderna vaccine, for which the original vaccine RNA sequence was not available at the time our study was carried out. Contigs encoding full-length spikes were assembled from the Moderna and Pfizer datasets. The Pfizer/BioNTech data [Figure 1] verified the reported sequence for that vaccine (https://berthub.eu/articles/posts/reverse-engineering-source-code-of-the-biontech-pfizer-vaccine/), while the Moderna sequence [Figure 2] could not be checked against a published reference. RNA preparations lacking dsRNA are desirable in generating vaccine formulations as these will minimize an otherwise dramatic biological (and nonspecific) response that vertebrates have to double stranded character in RNA (https://www.nature.com/articles/nrd.2017.243). In the sequence data that we analyzed, we found that the vast majority of reads were from the expected sense strand. In addition, the minority of antisense reads appeared different from sense reads in lacking the characteristic extensions expected from the template switching protocol. Examining only the reads with an evident template switch (as an indicator for strand-of-origin), we observed that both vaccines overwhelmingly yielded sense reads (>99.99%). Independent sequencing assays and other experimental measurements are ongoing and will be needed to determine whether this template-switched sense read fraction in the SmarterSeq protocol indeed represents the actual dsRNA content in the original material. This work provides an initial assessment of two RNAs that are now a part of the human ecosystem and that are likely to appear in numerous other high throughput RNA-seq studies in which a fraction of the individuals may have previously been vaccinated. ProtoAcknowledgements: Thanks to our colleagues for help and suggestions (Nimit Jain, Emily Greenwald, Lamia Wahba, William Wang, Amisha Kumar, Sameer Sundrani, David Lipman, Bijoyita Roy). Figure 1: Spike-encoding contig assembled from BioNTech/Pfizer BNT-162b2 vaccine. Although the full coding region is included, the nature of the methodology used for sequencing and assembly is such that the assembled contig could lack some sequence from the ends of the RNA. Within the assembled sequence, this hypothetical sequence shows a perfect match to the corresponding sequence from documents available online derived from manufacturer communications with the World Health Organization [as reported by https://berthub.eu/articles/posts/reverse-engineering-source-code-of-the-biontech-pfizer-vaccine/]. The 5ā end for the assembly matches the start site noted in these documents, while the read-based assembly lacks an interrupted polyA tail (A30(GCATATGACT)A70) that is expected to be present in the mRNA.
chatbot-webapp-boilerplate
metavoice-src
AI for human-level speech intelligence
react-leaflet-markercluster
React wrapper of the official Leaflet.markercluster for react-leaflet 4
threejs-react-configurator
vllm
A high-throughput and memory-efficient inference and serving engine for LLMs
l4b4r4b4b4's Repositories
l4b4r4b4b4/AIDocks
LLM-Training-API: Including Embeddings & ReRankers, mergekit, LaserRMT
l4b4r4b4b4/vllm
A high-throughput and memory-efficient inference and serving engine for LLMs
l4b4r4b4b4/chatbot-webapp-boilerplate
l4b4r4b4b4/metavoice-src
AI for human-level speech intelligence
l4b4r4b4b4/Agentless
Agentlessš±: an agentless approach to automatically solve software development problems
l4b4r4b4b4/arxiv-rss-assistant
l4b4r4b4b4/AutoArch
Automated Arch Linux Install
l4b4r4b4b4/chroma
the AI-native open-source embedding database
l4b4r4b4b4/docs
l4b4r4b4b4/dora
Implementation of DoRA
l4b4r4b4b4/dspy
DSPy: The framework for programmingānot promptingāfoundation models
l4b4r4b4b4/entity-recognition-datasets
A collection of corpora for named entity recognition (NER) and entity recognition tasks. These annotated datasets cover a variety of languages, domains and entity types.
l4b4r4b4b4/functionary
Chat language model that can use tools and interpret the results
l4b4r4b4b4/generative-models
Generative Models by Stability AI
l4b4r4b4b4/go-chroma
Go port of Chroma vector storage
l4b4r4b4b4/go-http-service-template
l4b4r4b4b4/go-restapi
l4b4r4b4b4/godot4-ai-npc-example
A Godot 4 Artificial Intelligence (AI) NPC example project.
l4b4r4b4b4/langgraphjs
ā” Build language agents as graphs ā”
l4b4r4b4b4/laserRMT
This is our own implementation of 'Layer Selective Rank Reduction'
l4b4r4b4b4/LocalAI
:robot: The free, Open Source OpenAI alternative. Self-hosted, community-driven and local-first. Drop-in replacement for OpenAI running on consumer-grade hardware. No GPU required. Runs ggml, gguf, GPTQ, onnx, TF compatible models: llama, llama2, rwkv, whisper, vicuna, koala, cerebras, falcon, dolly, starcoder, and many others
l4b4r4b4b4/multinerd
Repository for the paper "MultiNERD: A Multilingual, Multi-Genre and Fine-Grained Dataset for Named Entity Recognition (and Disambiguation)" (NAACL 2022).
l4b4r4b4b4/next-auth
Authentication for the Web.
l4b4r4b4b4/outlines
Structured Text Generation
l4b4r4b4b4/Real-Time-Latent-Consistency-Model
Demo showcasing ~real-time Latent Consistency Model pipeline with Diffusers and a MJPEG stream server
l4b4r4b4b4/SparsePrimingRepresentations
Public repo to document some SPR stuff
l4b4r4b4b4/tennis_analysis
This project analyzes Tennis players in a video to measure their speed, ball shot speed and number of shots. This project will detect players and the tennis ball using YOLO and also utilizes CNNs to extract court keypoints. This hands on project is perfect for polishing your machine learning, and computer vision skills.
l4b4r4b4b4/transformers
š¤ Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
l4b4r4b4b4/trl
Train transformer language models with reinforcement learning.
l4b4r4b4b4/unsloth
Finetune Llama 3.1, Mistral, Phi & Gemma LLMs 2-5x faster with 80% less memory